guides: fix the cross-reference direction in SEAT-IDENTITY

rev-code-01's non-blocking nit on #1313 round 2. The no-linking-step paragraph
pointed at the bridge explanation as 'described below'; it is above. Now names the
section, which survives further reordering better than a direction word does.

Text-only. Verified with the repo's PINNED prettier (3.8.1 via pnpm-lock.yaml) and
the sanitization gate, both clean.

README.md

@@ -138,9 +138,9 @@ mosaic brain tasks
 mosaic brain conversations
 
 # Agent forge pipeline
-mosaic forge run
+mosaic forge run [--simulate]        # fails closed (FORGE_NO_EXECUTOR) with no executor wired; --simulate for typed simulated runs
 mosaic forge status
-mosaic forge resume
+mosaic forge resume [--simulate]     # same fail-closed rule as forge run
 mosaic forge personas
 
 # Structured logging

apps/web/src/components/freshness/freshness-notices.tsx

@@ -0,0 +1,110 @@
+'use client';
+
+import type { ReactElement } from 'react';
+import { formatAge, type FreshnessLabel } from '@/lib/freshness/model';
+
+/**
+ * Rendering rules for non-current freshness states (RI-5-001).
+ *
+ * - `unavailable` renders an explicit failure panel — never an empty
+ *   healthy collection.
+ * - `stale` may render last-known data, but only under a visible label
+ *   carrying source identity, snapshot version, and age.
+ * - `partial` renders the verified parts plus an explicit list of what is
+ *   missing.
+ */
+
+interface RetryableNoticeProps {
+  readonly onRetry?: () => void;
+  readonly retryLabel?: string;
+}
+
+function RetryButton({ onRetry, retryLabel }: RetryableNoticeProps): ReactElement | null {
+  if (!onRetry) return null;
+  return (
+    <button
+      type="button"
+      onClick={onRetry}
+      className="mt-2 rounded-lg border border-surface-border px-3 py-1.5 text-xs transition-colors hover:border-gray-500"
+    >
+      {retryLabel ?? 'Retry'}
+    </button>
+  );
+}
+
+export interface UnavailableDataNoticeProps extends RetryableNoticeProps {
+  /** What is unavailable, e.g. "Tasks". */
+  readonly title: string;
+  /** Optional underlying failure detail (network message, invalidation reason). */
+  readonly detail?: string | null;
+}
+
+/** Explicit `unavailable` state. Never renders as an empty healthy collection. */
+export function UnavailableDataNotice({
+  title,
+  detail,
+  onRetry,
+  retryLabel,
+}: UnavailableDataNoticeProps): ReactElement {
+  return (
+    <div role="alert" className="rounded-lg border border-error/40 px-4 py-3 text-sm">
+      <p className="font-medium text-text-primary">{title} are unavailable</p>
+      <p className="mt-1 text-text-muted">
+        This is not an empty result — the data could not be verified from the gateway.
+        {detail ? ` ${detail}` : ''}
+      </p>
+      <RetryButton onRetry={onRetry} retryLabel={retryLabel} />
+    </div>
+  );
+}
+
+export interface StaleDataNoticeProps extends RetryableNoticeProps {
+  /** Provenance of the last-known snapshot being displayed. */
+  readonly label: FreshnessLabel;
+}
+
+/**
+ * Situational-awareness banner for `stale` data: last-known data may render,
+ * but visibly labeled with source identity, snapshot version, and age.
+ */
+export function StaleDataNotice({
+  label,
+  onRetry,
+  retryLabel,
+}: StaleDataNoticeProps): ReactElement {
+  return (
+    <div role="status" className="rounded-lg border border-warning/40 px-4 py-3 text-sm">
+      <p className="font-medium text-warning">Showing last-known data — it may be out of date</p>
+      <p className="mt-1 text-xs text-text-muted">
+        Source {label.source} · snapshot v{label.version} · fetched{' '}
+        {formatAge(label.fetchedAt, Date.now())}. Verdicts derived from this data are unknown and
+        changes are disabled until it is revalidated.
+      </p>
+      <RetryButton onRetry={onRetry} retryLabel={retryLabel ?? 'Revalidate'} />
+    </div>
+  );
+}
+
+export interface PartialDataNoticeProps extends RetryableNoticeProps {
+  /** Display names of the sections whose collections are unavailable. */
+  readonly missing: readonly string[];
+}
+
+/** `partial` surface banner: verified parts render, missing parts are explicit. */
+export function PartialDataNotice({
+  missing,
+  onRetry,
+  retryLabel,
+}: PartialDataNoticeProps): ReactElement {
+  return (
+    <div role="status" className="rounded-lg border border-warning/40 px-4 py-3 text-sm">
+      <p className="font-medium text-warning">Some data could not be loaded</p>
+      <p className="mt-1 text-xs text-text-muted">
+        {missing.join(', ')} {missing.length === 1 ? 'is' : 'are'} unavailable — sections below show
+        an explicit unavailable state instead of an empty list. Derived verdicts remain unknown
+        until every collection is revalidated.
+      </p>
+      <RetryButton onRetry={onRetry} retryLabel={retryLabel ?? 'Revalidate'} />
+    </div>
+  );
+}

apps/web/src/lib/freshness/model.spec.ts

@@ -0,0 +1,324 @@
+import { describe, expect, it } from 'vitest';
+import type { Task } from '@/lib/types';
+import {
+  acceptSnapshot,
+  assertMutable,
+  canMutate,
+  combineFreshness,
+  computeDigest,
+  computeFreshness,
+  DEFAULT_FRESHNESS_POLICY,
+  formatAge,
+  type FreshSnapshot,
+  invalidationReasonLabels,
+  StaleMutationError,
+  UNKNOWN_VERDICT,
+  verdictValue,
+} from './model';
+import { validateProjectCollection, validateTaskCollection } from './validators';
+
+const NOW = 1_800_000_000_000;
+
+const policy = { ...DEFAULT_FRESHNESS_POLICY, staleAfterMs: 60_000 };
+
+const taskPayload: Task[] = [
+  {
+    id: 'task-1',
+    title: 'T1',
+    description: null,
+    status: 'not-started',
+    priority: 'high',
+    projectId: 'project-1',
+    missionId: null,
+    assignee: null,
+    tags: null,
+    dueDate: null,
+    metadata: null,
+    createdAt: '2026-08-01T00:00:00.000Z',
+    updatedAt: '2026-08-01T00:00:00.000Z',
+  },
+];
+
+function acceptedTaskSnapshot(
+  overrides: Partial<FreshSnapshot<typeof taskPayload>> = {},
+): FreshSnapshot<typeof taskPayload> {
+  const result = acceptSnapshot({
+    value: taskPayload,
+    validate: validateTaskCollection,
+    previous: null,
+    policy,
+    source: 'gateway:/api/tasks',
+    now: NOW,
+  });
+  if (result.outcome !== 'accepted') {
+    throw new Error(`fixture setup failed: ${result.reason}`);
+  }
+  return { ...result.snapshot, ...overrides };
+}
+
+describe('computeFreshness', () => {
+  it('treats a missing snapshot as unavailable, never as an empty healthy collection', () => {
+    expect(computeFreshness({ snapshot: null, policy, now: NOW })).toBe('unavailable');
+  });
+
+  it('returns current for a fresh verified snapshot regardless of data emptiness', () => {
+    const empty = acceptSnapshot({
+      value: [],
+      validate: validateTaskCollection,
+      previous: null,
+      policy,
+      source: 'gateway:/api/tasks',
+      now: NOW,
+    });
+    if (empty.outcome !== 'accepted') throw new Error('expected acceptance');
+    expect(computeFreshness({ snapshot: empty.snapshot, policy, now: NOW })).toBe('current');
+  });
+
+  it('degrades to stale once the snapshot ages past staleAfterMs', () => {
+    const snapshot = acceptedTaskSnapshot();
+    expect(computeFreshness({ snapshot, policy, now: NOW + 60_001 })).toBe('stale');
+    expect(computeFreshness({ snapshot, policy, now: NOW + 59_999 })).toBe('current');
+  });
+
+  it('degrades to stale when the latest revalidation failed', () => {
+    const snapshot = acceptedTaskSnapshot();
+    expect(computeFreshness({ snapshot, policy, now: NOW, degraded: true })).toBe('stale');
+  });
+});
+
+describe('mutation guard', () => {
+  it('permits mutations only on current data', () => {
+    expect(canMutate('current')).toBe(true);
+    for (const state of ['stale', 'partial', 'unknown', 'unavailable'] as const) {
+      expect(canMutate(state)).toBe(false);
+    }
+  });
+
+  it('refuses mutations on non-current data via assertMutable', () => {
+    expect(() => assertMutable('current')).not.toThrow();
+    for (const state of ['stale', 'partial', 'unknown', 'unavailable'] as const) {
+      let thrown: unknown;
+      try {
+        assertMutable(state);
+      } catch (caught) {
+        thrown = caught;
+      }
+      expect(thrown).toBeInstanceOf(StaleMutationError);
+      expect(thrown).toBeInstanceOf(Error);
+      if (thrown instanceof StaleMutationError) {
+        expect(thrown.name).toBe('StaleMutationError');
+        expect(thrown.freshness).toBe(state);
+        expect(thrown.message).toContain(state);
+        expect(thrown.message).toContain('revalidat');
+      }
+    }
+  });
+});
+
+describe('acceptSnapshot', () => {
+  it('accepts a valid payload with provenance', () => {
+    const result = acceptSnapshot({
+      value: taskPayload,
+      validate: validateTaskCollection,
+      previous: null,
+      policy,
+      source: 'gateway:/api/tasks',
+      now: NOW,
+    });
+    expect(result.outcome).toBe('accepted');
+    if (result.outcome !== 'accepted') return;
+    expect(result.snapshot.source).toBe('gateway:/api/tasks');
+    expect(result.snapshot.version).toBe(1);
+    expect(result.snapshot.fetchedAt).toBe(NOW);
+    expect(result.snapshot.data).toEqual(taskPayload);
+  });
+
+  it('invalidates a schema-mismatched payload instead of rendering it', () => {
+    const result = acceptSnapshot({
+      value: { not: 'an array' },
+      validate: validateTaskCollection,
+      previous: acceptedTaskSnapshot(),
+      policy,
+      source: 'gateway:/api/tasks',
+      now: NOW,
+    });
+    expect(result).toEqual({ outcome: 'invalidated', reason: 'schema-mismatch' });
+    expect(invalidationReasonLabels['schema-mismatch']).toContain('schema');
+  });
+
+  it('invalidates cross-workspace payloads', () => {
+    const userOne = acceptSnapshot({
+      value: [
+        {
+          id: 'p1',
+          name: 'P1',
+          description: null,
+          status: 'active',
+          userId: 'user-1',
+          metadata: null,
+          createdAt: '2026-08-01T00:00:00.000Z',
+          updatedAt: '2026-08-01T00:00:00.000Z',
+        },
+      ],
+      validate: validateProjectCollection,
+      previous: null,
+      policy,
+      source: 'gateway:/api/projects',
+      now: NOW,
+    });
+    if (userOne.outcome !== 'accepted') throw new Error('expected acceptance');
+
+    const switched = acceptSnapshot({
+      value: [
+        {
+          id: 'p9',
+          name: 'P9',
+          description: null,
+          status: 'active',
+          userId: 'user-2',
+          metadata: null,
+          createdAt: '2026-08-01T00:00:00.000Z',
+          updatedAt: '2026-08-01T00:00:00.000Z',
+        },
+      ],
+      validate: validateProjectCollection,
+      previous: userOne.snapshot,
+      policy,
+      source: 'gateway:/api/projects',
+      now: NOW,
+    });
+    expect(switched).toEqual({ outcome: 'invalidated', reason: 'cross-workspace' });
+  });
+
+  it('keeps the previous workspace for collections with no intrinsic identity', () => {
+    const userOne = acceptSnapshot({
+      value: [
+        {
+          id: 'p1',
+          name: 'P1',
+          description: null,
+          status: 'active',
+          userId: 'user-1',
+          metadata: null,
+          createdAt: '2026-08-01T00:00:00.000Z',
+          updatedAt: '2026-08-01T00:00:00.000Z',
+        },
+      ],
+      validate: validateProjectCollection,
+      previous: null,
+      policy,
+      source: 'gateway:/api/projects',
+      now: NOW,
+    });
+    if (userOne.outcome !== 'accepted') throw new Error('expected acceptance');
+
+    // Empty list after the user deleted every project: no identity to check,
+    // so the verified scope is retained and the empty state stays healthy.
+    const emptied = acceptSnapshot({
+      value: [],
+      validate: validateProjectCollection,
+      previous: userOne.snapshot,
+      policy,
+      source: 'gateway:/api/projects',
+      now: NOW,
+    });
+    expect(emptied.outcome).toBe('accepted');
+    if (emptied.outcome === 'accepted') {
+      expect(emptied.snapshot.data).toEqual([]);
+      expect(emptied.snapshot.workspace).toBe('user-1');
+    }
+  });
+
+  it('invalidates version regressions', () => {
+    const previous = acceptedTaskSnapshot({ version: 7 });
+    const regressed = acceptSnapshot({
+      value: taskPayload,
+      validate: validateTaskCollection,
+      previous,
+      policy,
+      source: 'gateway:/api/tasks',
+      now: NOW,
+      incomingVersion: 3,
+    });
+    expect(regressed).toEqual({ outcome: 'invalidated', reason: 'version-regression' });
+
+    const newerSchema = acceptedTaskSnapshot({ schemaVersion: 4 });
+    const downgradedClient = acceptSnapshot({
+      value: taskPayload,
+      validate: validateTaskCollection,
+      previous: newerSchema,
+      policy: { ...policy, schemaVersion: 2 },
+      source: 'gateway:/api/tasks',
+      now: NOW,
+    });
+    expect(downgradedClient).toEqual({ outcome: 'invalidated', reason: 'version-regression' });
+  });
+
+  it('increments the version monotonically across accepted snapshots', () => {
+    const first = acceptedTaskSnapshot();
+    const second = acceptSnapshot({
+      value: taskPayload,
+      validate: validateTaskCollection,
+      previous: first,
+      policy,
+      source: 'gateway:/api/tasks',
+      now: NOW,
+    });
+    expect(second.outcome).toBe('accepted');
+    if (second.outcome === 'accepted') {
+      expect(second.snapshot.version).toBe(first.version + 1);
+    }
+  });
+});
+
+describe('combineFreshness', () => {
+  it('gates the surface on the primary collection', () => {
+    expect(combineFreshness('unavailable', ['current'])).toBe('unavailable');
+    expect(combineFreshness('unknown', ['current'])).toBe('unknown');
+    expect(combineFreshness('current', [])).toBe('current');
+  });
+
+  it('degrades to partial when a secondary is unavailable', () => {
+    expect(combineFreshness('current', ['current', 'unavailable'])).toBe('partial');
+  });
+
+  it('degrades to unknown while a secondary is still loading', () => {
+    expect(combineFreshness('current', ['unknown'])).toBe('unknown');
+  });
+
+  it('degrades to stale when any collection is stale', () => {
+    expect(combineFreshness('current', ['stale'])).toBe('stale');
+    expect(combineFreshness('stale', ['current'])).toBe('stale');
+  });
+
+  it('propagates partial secondaries', () => {
+    expect(combineFreshness('current', ['partial'])).toBe('partial');
+  });
+});
+
+describe('computeDigest', () => {
+  it('is stable across key order and changes with data', () => {
+    const a = computeDigest({ x: 1, y: [1, 2] });
+    const b = computeDigest({ y: [1, 2], x: 1 });
+    expect(a).toBe(b);
+    expect(computeDigest({ x: 1, y: [1, 3] })).not.toBe(a);
+  });
+});
+
+describe('verdictValue', () => {
+  it('returns the value only for verified inputs', () => {
+    expect(verdictValue(true, '5')).toBe('5');
+    expect(verdictValue(false, '5')).toBe(UNKNOWN_VERDICT);
+    expect(verdictValue(false, '5')).not.toBe('5');
+  });
+});
+
+describe('formatAge', () => {
+  it('labels age in human terms', () => {
+    expect(formatAge(NOW, NOW)).toBe('just now');
+    expect(formatAge(NOW, NOW + 15_000)).toBe('under a minute ago');
+    expect(formatAge(NOW, NOW + 120_000)).toBe('2m ago');
+    expect(formatAge(NOW, NOW + 3 * 3_600_000)).toBe('3h ago');
+    expect(formatAge(NOW, NOW + 2 * 86_400_000)).toBe('2d ago');
+  });
+});

apps/web/src/lib/freshness/model.ts

@@ -0,0 +1,261 @@
+/**
+ * Typed freshness model for gateway-fetched collections (RI-5-001).
+ *
+ * A failed or stale fetch must never be indistinguishable from an empty
+ * healthy collection. Every fetched surface carries an explicit freshness
+ * state, a verified snapshot identity (source, workspace, version, age), and
+ * a mutation guard that refuses state-changing operations unless the data is
+ * verified current.
+ */
+
+/** Freshness states for fetched data. Never inferred from emptiness. */
+export type FreshnessState = 'current' | 'stale' | 'partial' | 'unknown' | 'unavailable';
+
+/**
+ * Reasons a snapshot is invalidated. An invalidated snapshot is treated as
+ * unavailable and is never rendered as current.
+ */
+export type InvalidationReason =
+  | 'cache-corruption'
+  | 'cross-workspace'
+  | 'schema-mismatch'
+  | 'version-regression';
+
+/** Human-readable labels for invalidation reasons (UI + error messages). */
+export const invalidationReasonLabels: Record<InvalidationReason, string> = {
+  'cache-corruption': 'cached snapshot failed integrity checks',
+  'cross-workspace': 'data belongs to a different workspace',
+  'schema-mismatch': 'response did not match the expected schema',
+  'version-regression': 'snapshot version regressed below the accepted version',
+};
+
+/** A verified snapshot of fetched data with full provenance. */
+export interface FreshSnapshot<T> {
+  readonly data: T;
+  /** Source identity of the fetch, e.g. `gateway:/api/tasks`. */
+  readonly source: string;
+  /** Workspace scope the data belongs to. */
+  readonly workspace: string;
+  /** Monotonic snapshot sequence number for this surface. */
+  readonly version: number;
+  /** Schema version of the validator that accepted this snapshot. */
+  readonly schemaVersion: number;
+  /** Epoch ms at which the data was verified. */
+  readonly fetchedAt: number;
+  /** Integrity digest of `data`, used to detect cache corruption. */
+  readonly digest: string;
+}
+
+/** Provenance label rendered next to last-known data. */
+export interface FreshnessLabel {
+  readonly source: string;
+  readonly version: number;
+  readonly fetchedAt: number;
+}
+
+/** Policy governing freshness for a surface. */
+export interface FreshnessPolicy {
+  /** Active workspace scope. Snapshots from other scopes are invalidated. */
+  readonly workspace: string;
+  /** Schema version of the current validator. */
+  readonly schemaVersion: number;
+  /** Age after which a verified snapshot degrades from current to stale. */
+  readonly staleAfterMs: number;
+}
+
+export const DEFAULT_FRESHNESS_POLICY: FreshnessPolicy = {
+  workspace: 'default',
+  schemaVersion: 1,
+  staleAfterMs: 60_000,
+};
+
+/** Payload returned by a successful schema validation. */
+export interface FreshPayload<T> {
+  readonly data: T;
+  /**
+   * Workspace identity extracted from the payload itself when the collection
+   * carries one (e.g. a uniform `userId` on projects). `null` when the
+   * collection has no intrinsic workspace identity.
+   */
+  readonly workspace: string | null;
+}
+
+/** Error thrown when a mutation is attempted on non-current data. */
+export class StaleMutationError extends Error {
+  readonly freshness: FreshnessState;
+
+  constructor(freshness: FreshnessState) {
+    super(`Refused mutation on ${freshness} data: revalidation is required before mutating.`);
+    this.name = 'StaleMutationError';
+    this.freshness = freshness;
+  }
+}
+
+/** Stable JSON digest used for snapshot integrity checks. */
+export function computeDigest(value: unknown): string {
+  // FNV-1a 32-bit over the stable JSON serialization. This is an integrity
+  // check against corruption, not a cryptographic guarantee.
+  let hash = 0x811c9dc5;
+  for (const byte of stableStringify(value)) {
+    hash ^= byte.charCodeAt(0);
+    hash = Math.imul(hash, 0x01000193) >>> 0;
+  }
+  return hash.toString(16).padStart(8, '0');
+}
+
+function stableStringify(value: unknown): string {
+  return serialize(value);
+}
+
+function serialize(value: unknown): string {
+  if (value === null || typeof value !== 'object') return JSON.stringify(value) ?? 'null';
+  if (Array.isArray(value)) return `[${value.map(serialize).join(',')}]`;
+  const entries = Object.entries(value as Record<string, unknown>)
+    .filter(([, item]) => item !== undefined)
+    .sort(([left], [right]) => (left < right ? -1 : left > right ? 1 : 0))
+    .map(([key, item]) => `${JSON.stringify(key)}:${serialize(item)}`);
+  return `{${entries.join(',')}}`;
+}
+
+export type AcceptSnapshotResult<T> =
+  | { readonly outcome: 'accepted'; readonly snapshot: FreshSnapshot<T> }
+  | { readonly outcome: 'invalidated'; readonly reason: InvalidationReason };
+
+export interface AcceptSnapshotOptions<T> {
+  /** Raw fetched value (untrusted JSON). */
+  readonly value: unknown;
+  /** Schema validator; returns `null` when the value does not match. */
+  readonly validate: (value: unknown) => FreshPayload<T> | null;
+  /** Previously accepted snapshot for this surface, if any. */
+  readonly previous: FreshSnapshot<T> | null;
+  readonly policy: FreshnessPolicy;
+  readonly source: string;
+  /**
+   * Version carried by the incoming payload when the transport exposes one.
+   * Must not regress below the accepted snapshot's version.
+   */
+  readonly incomingVersion?: number;
+  readonly now: number;
+}
+
+/**
+ * Validate and accept a fetched value as a snapshot, or invalidate it.
+ *
+ * Invalidation rules (each treated as unavailable, never rendered current):
+ * - schema mismatch: the payload fails validation
+ * - cross-workspace: the payload's workspace differs from the verified one
+ * - version regression: payload/schema version is below the accepted one
+ */
+export function acceptSnapshot<T>(options: AcceptSnapshotOptions<T>): AcceptSnapshotResult<T> {
+  const payload = options.validate(options.value);
+  if (payload === null) {
+    return { outcome: 'invalidated', reason: 'schema-mismatch' };
+  }
+
+  // Workspace identity: the payload's own scope wins; a collection with no
+  // intrinsic identity (e.g. an empty list after every project was deleted)
+  // keeps the previously verified scope rather than resetting to the policy
+  // default, so a legitimately empty response is not mistaken for a scope
+  // change.
+  const workspace = payload.workspace ?? options.previous?.workspace ?? options.policy.workspace;
+  if (options.previous !== null && options.previous.workspace !== workspace) {
+    return { outcome: 'invalidated', reason: 'cross-workspace' };
+  }
+  if (options.previous !== null && options.policy.schemaVersion < options.previous.schemaVersion) {
+    return { outcome: 'invalidated', reason: 'version-regression' };
+  }
+  if (
+    options.incomingVersion !== undefined &&
+    options.previous !== null &&
+    options.incomingVersion < options.previous.version
+  ) {
+    return { outcome: 'invalidated', reason: 'version-regression' };
+  }
+
+  const snapshot: FreshSnapshot<T> = {
+    data: payload.data,
+    source: options.source,
+    workspace,
+    version: options.incomingVersion ?? (options.previous?.version ?? 0) + 1,
+    schemaVersion: options.policy.schemaVersion,
+    fetchedAt: options.now,
+    digest: computeDigest(payload.data),
+  };
+  return { outcome: 'accepted', snapshot };
+}
+
+export interface ComputeFreshnessOptions {
+  readonly snapshot: FreshSnapshot<unknown> | null;
+  readonly policy: FreshnessPolicy;
+  readonly now: number;
+  /**
+   * True when the snapshot cannot be trusted as current regardless of age:
+   * the latest revalidation failed, or the snapshot was restored from cache
+   * and has not been verified by a fetch in this session.
+   */
+  readonly degraded?: boolean;
+}
+
+/**
+ * Compute the freshness state of a snapshot. A missing snapshot is
+ * `unavailable` (never "empty and healthy"); a degraded or aged snapshot is
+ * `stale` (situational awareness only).
+ */
+export function computeFreshness(options: ComputeFreshnessOptions): FreshnessState {
+  const { snapshot, policy, now, degraded = false } = options;
+  if (snapshot === null) return 'unavailable';
+  if (degraded) return 'stale';
+  if (now - snapshot.fetchedAt > policy.staleAfterMs) return 'stale';
+  return 'current';
+}
+
+/** Only verified-current data may back a state-changing action. */
+export function canMutate(state: FreshnessState): boolean {
+  return state === 'current';
+}
+
+/** Defense in depth: reject the mutation call itself on non-current data. */
+export function assertMutable(state: FreshnessState): void {
+  if (!canMutate(state)) {
+    throw new StaleMutationError(state);
+  }
+}
+
+/**
+ * Combine freshness across a multi-collection surface (primary + secondaries).
+ * The primary collection gates the surface: unknown while it loads,
+ * unavailable when it fails. Missing secondaries degrade the surface to
+ * `partial`; aged collections degrade it to `stale`.
+ */
+export function combineFreshness(
+  primary: FreshnessState,
+  secondaries: readonly FreshnessState[],
+): FreshnessState {
+  if (primary === 'unavailable') return 'unavailable';
+  if (primary === 'unknown') return 'unknown';
+  if (secondaries.includes('unavailable')) return 'partial';
+  if (secondaries.includes('unknown')) return 'unknown';
+  if (secondaries.includes('stale') || primary === 'stale') return 'stale';
+  if (secondaries.includes('partial')) return 'partial';
+  return 'current';
+}
+
+/** Render-safe age label for snapshot provenance. */
+export function formatAge(fetchedAt: number, now: number): string {
+  const ageMs = Math.max(0, now - fetchedAt);
+  if (ageMs < 10_000) return 'just now';
+  const minutes = Math.floor(ageMs / 60_000);
+  if (minutes < 1) return 'under a minute ago';
+  if (minutes < 60) return `${minutes}m ago`;
+  const hours = Math.floor(minutes / 60);
+  if (hours < 24) return `${hours}h ago`;
+  const days = Math.floor(hours / 24);
+  return `${days}d ago`;
+}
+
+/** Derived verdict placeholder for non-current inputs — never a green value. */
+export const UNKNOWN_VERDICT = '?';
+
+export function verdictValue(verified: boolean, value: string): string {
+  return verified ? value : UNKNOWN_VERDICT;
+}

apps/web/src/lib/freshness/snapshot-cache.spec.ts

@@ -0,0 +1,197 @@
+import { afterEach, beforeEach, describe, expect, it } from 'vitest';
+import { acceptSnapshot, DEFAULT_FRESHNESS_POLICY } from './model';
+import { clearSnapshotCache, readSnapshotCache, writeSnapshotCache } from './snapshot-cache';
+import { validateProjectCollection, validateTaskCollection } from './validators';
+import { projectFixtures, taskFixtures } from '@/spa/pages/page-fixtures';
+import type { Project, Task } from '@/lib/types';
+
+const KEY = 'test:tasks';
+const NOW = 1_800_000_000_000;
+const policy = { ...DEFAULT_FRESHNESS_POLICY, staleAfterMs: 60_000 };
+
+function storedTaskSnapshot() {
+  const result = acceptSnapshot({
+    value: taskFixtures,
+    validate: validateTaskCollection,
+    previous: null,
+    policy,
+    source: 'gateway:/api/tasks',
+    now: NOW,
+  });
+  if (result.outcome !== 'accepted') throw new Error('fixture setup failed');
+  return result.snapshot;
+}
+
+function storedProjectSnapshot() {
+  const result = acceptSnapshot({
+    value: projectFixtures,
+    validate: validateProjectCollection,
+    previous: null,
+    policy,
+    source: 'gateway:/api/projects',
+    now: NOW,
+  });
+  if (result.outcome !== 'accepted') throw new Error('fixture setup failed');
+  return result.snapshot;
+}
+
+function readTasks() {
+  return readSnapshotCache({
+    key: KEY,
+    workspace: policy.workspace,
+    policy,
+    validate: validateTaskCollection,
+  });
+}
+
+/** Write an arbitrary value directly at the raw cache slot. */
+function writeRaw(key: string, value: unknown): void {
+  sessionStorage.setItem(`mosaic:freshness:v1:${key}`, JSON.stringify(value));
+}
+
+/** Parse and re-write the stored entry (for tampering with internals). */
+function tamperStored<T>(key: string, mutate: (stored: T) => void): void {
+  const parsed = JSON.parse(sessionStorage.getItem(`mosaic:freshness:v1:${key}`) ?? '{}') as T;
+  mutate(parsed);
+  writeRaw(key, parsed);
+}
+
+beforeEach(() => {
+  sessionStorage.clear();
+});
+
+afterEach(() => {
+  sessionStorage.clear();
+});
+
+describe('readSnapshotCache', () => {
+  it('misses when nothing is stored', () => {
+    expect(readTasks()).toEqual({ outcome: 'miss' });
+  });
+
+  it('hits for a well-formed entry and preserves provenance', () => {
+    const snapshot = storedTaskSnapshot();
+    writeSnapshotCache(KEY, snapshot);
+
+    const result = readTasks();
+    expect(result.outcome).toBe('hit');
+    if (result.outcome === 'hit') {
+      expect(result.snapshot.data).toEqual(taskFixtures);
+      expect(result.snapshot.source).toBe('gateway:/api/tasks');
+      expect(result.snapshot.version).toBe(snapshot.version);
+      expect(result.snapshot.fetchedAt).toBe(snapshot.fetchedAt);
+      expect(result.snapshot.workspace).toBe(snapshot.workspace);
+    }
+  });
+
+  it('invalidates unparsable entries as cache corruption', () => {
+    sessionStorage.setItem(`mosaic:freshness:v1:${KEY}`, '{not json');
+    expect(readTasks()).toEqual({ outcome: 'invalidated', reason: 'cache-corruption' });
+  });
+
+  it('invalidates structurally wrong entries as cache corruption', () => {
+    const malformed: unknown[] = [
+      'nested but not a snapshot',
+      { data: taskFixtures }, // missing provenance fields
+      {
+        data: taskFixtures,
+        source: 1,
+        workspace: 'w',
+        version: 1,
+        schemaVersion: 1,
+        fetchedAt: 1,
+        digest: 'x',
+      },
+      null,
+      17,
+    ];
+    for (const entry of malformed) {
+      writeRaw(KEY, entry);
+      expect(readTasks()).toEqual({ outcome: 'invalidated', reason: 'cache-corruption' });
+    }
+  });
+
+  it('invalidates digest mismatches as cache corruption (tampered data)', () => {
+    writeSnapshotCache(KEY, storedTaskSnapshot());
+    tamperStored<{ data: Task[] }>(KEY, (stored) => {
+      stored.data = [...stored.data, { ...stored.data[0]!, id: 'injected-task' }];
+    });
+    expect(readTasks()).toEqual({ outcome: 'invalidated', reason: 'cache-corruption' });
+  });
+
+  it('invalidates entries scoped to another workspace', () => {
+    const snapshot = storedTaskSnapshot();
+    writeSnapshotCache(KEY, { ...snapshot, workspace: 'someone-else' });
+    expect(readTasks()).toEqual({ outcome: 'invalidated', reason: 'cross-workspace' });
+  });
+
+  it('invalidates entries written by a newer schema as a version regression', () => {
+    const snapshot = storedTaskSnapshot();
+    writeSnapshotCache(KEY, { ...snapshot, schemaVersion: policy.schemaVersion + 1 });
+    expect(readTasks()).toEqual({ outcome: 'invalidated', reason: 'version-regression' });
+  });
+
+  it('invalidates entries whose data no longer validates (schema mismatch)', () => {
+    writeSnapshotCache(KEY, storedTaskSnapshot());
+    tamperStored<{ data: unknown }>(KEY, (stored) => {
+      stored.data = { malformed: true };
+    });
+    expect(readTasks()).toEqual({ outcome: 'invalidated', reason: 'schema-mismatch' });
+  });
+
+  it('never reports a corrupted raw entry as a hit (negative control)', () => {
+    for (const raw of ['{oops', 'null', '"string"', '[]', '12']) {
+      sessionStorage.setItem(`mosaic:freshness:v1:${KEY}`, raw);
+      const result = readTasks();
+      expect(result.outcome).not.toBe('hit');
+      expect(result.outcome).toBe('invalidated');
+    }
+  });
+
+  it('scopes project collections by their workspace identity', () => {
+    const snapshot = storedProjectSnapshot();
+    writeSnapshotCache('test:projects', snapshot);
+
+    const sameScope = readSnapshotCache({
+      key: 'test:projects',
+      workspace: 'user-1',
+      policy,
+      validate: validateProjectCollection,
+    });
+    expect(sameScope.outcome).toBe('hit');
+
+    const foreignScope = readSnapshotCache({
+      key: 'test:projects',
+      workspace: 'user-2',
+      policy,
+      validate: validateProjectCollection,
+    });
+    expect(foreignScope).toEqual({ outcome: 'invalidated', reason: 'cross-workspace' });
+  });
+});
+
+describe('writeSnapshotCache round-trip', () => {
+  it('round-trips an accepted project snapshot', () => {
+    const snapshot = storedProjectSnapshot();
+    writeSnapshotCache('test:projects', snapshot);
+    const result = readSnapshotCache({
+      key: 'test:projects',
+      workspace: snapshot.workspace,
+      policy,
+      validate: validateProjectCollection,
+    });
+    expect(result.outcome).toBe('hit');
+    if (result.outcome === 'hit') {
+      expect(result.snapshot.data).toEqual(projectFixtures as Project[]);
+    }
+  });
+});
+
+describe('clearSnapshotCache', () => {
+  it('drops the entry so the next read misses', () => {
+    writeSnapshotCache(KEY, storedTaskSnapshot());
+    expect(readTasks().outcome).toBe('hit');
+    clearSnapshotCache(KEY);
+    expect(readTasks()).toEqual({ outcome: 'miss' });
+  });
+});

apps/web/src/lib/freshness/snapshot-cache.ts

@@ -0,0 +1,154 @@
+import {
+  computeDigest,
+  type FreshPayload,
+  type FreshSnapshot,
+  type FreshnessPolicy,
+  type InvalidationReason,
+} from './model';
+
+/**
+ * Session-scoped last-known snapshot cache (RI-5-001).
+ *
+ * Restored snapshots are situational awareness only: they surface as `stale`
+ * until a fetch re-verifies them. A cache entry that is corrupted, belongs to
+ * another workspace, was written by a newer schema, or no longer validates is
+ * invalidated (treated as unavailable, never rendered as current).
+ */
+
+const CACHE_PREFIX = 'mosaic:freshness:v1';
+
+interface StoredSnapshot {
+  data: unknown;
+  source: string;
+  workspace: string;
+  version: number;
+  schemaVersion: number;
+  fetchedAt: number;
+  digest: string;
+}
+
+export type SnapshotCacheRead<T> =
+  | { readonly outcome: 'hit'; readonly snapshot: FreshSnapshot<T> }
+  | { readonly outcome: 'miss' }
+  | { readonly outcome: 'invalidated'; readonly reason: InvalidationReason };
+
+export interface ReadSnapshotCacheOptions<T> {
+  readonly key: string;
+  readonly workspace: string;
+  readonly policy: FreshnessPolicy;
+  readonly validate: (value: unknown) => FreshPayload<T> | null;
+}
+
+function cacheKey(key: string): string {
+  return `${CACHE_PREFIX}:${key}`;
+}
+
+function isStoredSnapshot(value: unknown): value is StoredSnapshot {
+  if (typeof value !== 'object' || value === null) return false;
+  const candidate = value as Record<string, unknown>;
+  return (
+    typeof candidate['data'] === 'object' &&
+    candidate['data'] !== null &&
+    typeof candidate['source'] === 'string' &&
+    typeof candidate['workspace'] === 'string' &&
+    typeof candidate['version'] === 'number' &&
+    typeof candidate['schemaVersion'] === 'number' &&
+    typeof candidate['fetchedAt'] === 'number' &&
+    typeof candidate['digest'] === 'string'
+  );
+}
+
+function getStorage(): Storage | null {
+  try {
+    return globalThis.sessionStorage ?? null;
+  } catch {
+    return null;
+  }
+}
+
+/**
+ * Restore a cached snapshot under the active workspace scope. Every failure
+ * mode maps to an explicit invalidation reason or a miss — never to data
+ * that renders as current.
+ */
+export function readSnapshotCache<T>(options: ReadSnapshotCacheOptions<T>): SnapshotCacheRead<T> {
+  const storage = getStorage();
+  if (storage === null) return { outcome: 'miss' };
+
+  let raw: string | null;
+  try {
+    raw = storage.getItem(cacheKey(options.key));
+  } catch {
+    return { outcome: 'miss' };
+  }
+  if (raw === null) return { outcome: 'miss' };
+
+  let parsed: unknown;
+  try {
+    parsed = JSON.parse(raw);
+  } catch {
+    return { outcome: 'invalidated', reason: 'cache-corruption' };
+  }
+  if (!isStoredSnapshot(parsed)) {
+    return { outcome: 'invalidated', reason: 'cache-corruption' };
+  }
+  if (parsed.workspace !== options.workspace) {
+    return { outcome: 'invalidated', reason: 'cross-workspace' };
+  }
+  if (parsed.schemaVersion > options.policy.schemaVersion) {
+    // Written by a newer build than the running client: version regression.
+    return { outcome: 'invalidated', reason: 'version-regression' };
+  }
+
+  const payload = options.validate(parsed.data);
+  if (payload === null) {
+    return { outcome: 'invalidated', reason: 'schema-mismatch' };
+  }
+  if (computeDigest(payload.data) !== parsed.digest) {
+    return { outcome: 'invalidated', reason: 'cache-corruption' };
+  }
+
+  return {
+    outcome: 'hit',
+    snapshot: {
+      data: payload.data,
+      source: parsed.source,
+      workspace: parsed.workspace,
+      version: parsed.version,
+      schemaVersion: parsed.schemaVersion,
+      fetchedAt: parsed.fetchedAt,
+      digest: parsed.digest,
+    },
+  };
+}
+
+/** Persist a verified snapshot. Failures are non-fatal (cache is best-effort). */
+export function writeSnapshotCache<T>(key: string, snapshot: FreshSnapshot<T>): void {
+  const storage = getStorage();
+  if (storage === null) return;
+  const stored: StoredSnapshot = {
+    data: snapshot.data,
+    source: snapshot.source,
+    workspace: snapshot.workspace,
+    version: snapshot.version,
+    schemaVersion: snapshot.schemaVersion,
+    fetchedAt: snapshot.fetchedAt,
+    digest: snapshot.digest,
+  };
+  try {
+    storage.setItem(cacheKey(key), JSON.stringify(stored));
+  } catch {
+    // Quota or serialization failures simply skip caching.
+  }
+}
+
+/** Drop a cached snapshot (used when a surface invalidates its cache entry). */
+export function clearSnapshotCache(key: string): void {
+  const storage = getStorage();
+  if (storage === null) return;
+  try {
+    storage.removeItem(cacheKey(key));
+  } catch {
+    // Ignorable: a wedged storage entry is detected as corruption on read.
+  }
+}

apps/web/src/lib/freshness/use-fresh-collection.spec.tsx

@@ -0,0 +1,372 @@
+import { act } from 'react';
+import { createRoot, type Root } from 'react-dom/client';
+import { afterEach, beforeAll, beforeEach, describe, expect, it, vi } from 'vitest';
+import type { Task } from '@/lib/types';
+import { acceptSnapshot, StaleMutationError, DEFAULT_FRESHNESS_POLICY } from './model';
+import type { FreshnessFailure } from './use-fresh-collection';
+import {
+  describeFailure,
+  useFreshCollection,
+  type FreshCollection,
+  type UseFreshCollectionOptions,
+} from './use-fresh-collection';
+import { validateProjectCollection, validateTaskCollection } from './validators';
+import { projectFixtures, taskFixtures } from '@/spa/pages/page-fixtures';
+
+/**
+ * Failure-matrix coverage for the freshness seam (RI-5-001): network failure,
+ * auth failure, malformed response, cache corruption, stale age, schema
+ * mismatch, cross-workspace, recovery, and stale-action rejection — with
+ * negative controls proving no case yields current data or an enabled
+ * mutation.
+ */
+
+const NOW = 1_800_000_000_000;
+
+interface Deferred<T> {
+  promise: Promise<T>;
+  resolve: (value: T) => void;
+  reject: (reason?: unknown) => void;
+}
+
+function createDeferred<T>(): Deferred<T> {
+  let resolve!: (value: T) => void;
+  let reject!: (reason?: unknown) => void;
+  const promise = new Promise<T>((res, rej) => {
+    resolve = res;
+    reject = rej;
+  });
+  return { promise, resolve, reject };
+}
+
+let root: Root | null = null;
+let container: HTMLDivElement;
+let latest: FreshCollection<Task[]> | null = null;
+
+function Probe({
+  options,
+}: {
+  options: UseFreshCollectionOptions<Task[]>;
+}): React.ReactElement | null {
+  latest = useFreshCollection<Task[]>(options);
+  return null;
+}
+
+beforeAll(() => {
+  Object.defineProperty(globalThis, 'IS_REACT_ACT_ENVIRONMENT', {
+    configurable: true,
+    value: true,
+  });
+});
+
+beforeEach(() => {
+  sessionStorage.clear();
+});
+
+afterEach(async () => {
+  await act(async () => {
+    root?.unmount();
+  });
+  document.body.replaceChildren();
+  root = null;
+  latest = null;
+  sessionStorage.clear();
+  vi.restoreAllMocks();
+});
+
+async function renderCollection(
+  options: UseFreshCollectionOptions<Task[]>,
+): Promise<FreshCollection<Task[]>> {
+  container = document.createElement('div');
+  document.body.append(container);
+  root = createRoot(container);
+  await act(async () => {
+    root?.render(<Probe options={options} />);
+  });
+  if (latest === null) throw new Error('hook did not run');
+  return latest;
+}
+
+function taskOptions(
+  overrides: Partial<UseFreshCollectionOptions<Task[]>> = {},
+): UseFreshCollectionOptions<Task[]> {
+  return {
+    source: 'gateway:/api/tasks',
+    fetcher: () => Promise.resolve(taskFixtures),
+    validate: validateTaskCollection,
+    cacheKey: 'tasks',
+    clock: () => NOW,
+    ...overrides,
+  };
+}
+
+function authError(statusCode: number): Error & { statusCode: number } {
+  return Object.assign(new Error(`Request failed with ${statusCode}`), { statusCode });
+}
+
+function seedCache(key: string): number {
+  const result = acceptSnapshot({
+    value: taskFixtures,
+    validate: validateTaskCollection,
+    previous: null,
+    policy: DEFAULT_FRESHNESS_POLICY,
+    source: 'gateway:/api/tasks',
+    now: NOW,
+  });
+  if (result.outcome !== 'accepted') throw new Error('fixture setup failed');
+  sessionStorage.setItem(`mosaic:freshness:v1:${key}`, JSON.stringify({ ...result.snapshot }));
+  return result.snapshot.version;
+}
+
+describe('useFreshCollection failure matrix', () => {
+  it('is unknown (not empty) while the first validation is in flight', async () => {
+    const deferred = createDeferred<Task[]>();
+    const collection = await renderCollection(taskOptions({ fetcher: () => deferred.promise }));
+
+    expect(collection.freshness).toBe('unknown');
+    expect(collection.validating).toBe(true);
+    expect(collection.data).toBeNull();
+    expect(collection.canMutate).toBe(false);
+
+    await act(async () => {
+      deferred.resolve(taskFixtures);
+      await deferred.promise;
+    });
+  });
+
+  it('becomes current with provenance after a verified fetch', async () => {
+    const collection = await renderCollection(taskOptions());
+
+    expect(collection.freshness).toBe('current');
+    expect(collection.data).toEqual(taskFixtures);
+    expect(collection.snapshot?.source).toBe('gateway:/api/tasks');
+    expect(collection.snapshot?.version).toBe(1);
+    expect(collection.failure).toBeNull();
+    expect(collection.canMutate).toBe(true);
+    // Verified snapshot is persisted for last-known restore.
+    expect(sessionStorage.getItem('mosaic:freshness:v1:tasks')).toBeTruthy();
+  });
+
+  it('treats a network failure as unavailable — never an empty healthy collection', async () => {
+    const collection = await renderCollection(
+      taskOptions({ fetcher: () => Promise.reject(new Error('network down')) }),
+    );
+
+    expect(collection.freshness).toBe('unavailable');
+    expect(collection.data).toBeNull();
+    expect(collection.failure).toEqual({ kind: 'fetch', message: 'network down' });
+    expect(collection.canMutate).toBe(false);
+    expect(describeFailure(collection.failure)).toBe('network down');
+  });
+
+  it('treats an auth failure as unavailable and drops the last-known snapshot', async () => {
+    let call = 0;
+    const collection = await renderCollection(
+      taskOptions({
+        fetcher: () => {
+          call += 1;
+          return call === 1 ? Promise.resolve(taskFixtures) : Promise.reject(authError(401));
+        },
+      }),
+    );
+    expect(collection.freshness).toBe('current');
+
+    await act(async () => {
+      await collection.revalidate();
+    });
+
+    expect(latest?.freshness).toBe('unavailable');
+    expect(latest?.data).toBeNull();
+    expect(latest?.failure?.kind).toBe('fetch');
+    // The previous user's data must not linger in the session cache.
+    expect(sessionStorage.getItem('mosaic:freshness:v1:tasks')).toBeNull();
+  });
+
+  it('invalidates a malformed response as a schema mismatch', async () => {
+    const collection = await renderCollection(
+      taskOptions({ fetcher: () => Promise.resolve({ malformed: true }) }),
+    );
+
+    expect(collection.freshness).toBe('unavailable');
+    expect(collection.data).toBeNull();
+    expect(collection.failure).toEqual({ kind: 'invalidated', reason: 'schema-mismatch' });
+    expect(collection.canMutate).toBe(false);
+  });
+
+  it('keeps the previous snapshot as labeled stale when a later payload mismatches', async () => {
+    let call = 0;
+    const collection = await renderCollection(
+      taskOptions({
+        fetcher: () => {
+          call += 1;
+          return call === 1 ? Promise.resolve(taskFixtures) : Promise.resolve('garbage');
+        },
+      }),
+    );
+    expect(collection.freshness).toBe('current');
+
+    await act(async () => {
+      await collection.revalidate();
+    });
+
+    expect(latest?.freshness).toBe('stale');
+    expect(latest?.data).toEqual(taskFixtures);
+    expect(latest?.failure).toEqual({ kind: 'invalidated', reason: 'schema-mismatch' });
+    expect(latest?.canMutate).toBe(false);
+  });
+
+  it('drops the snapshot when the workspace changes under it (cross-workspace)', async () => {
+    let call = 0;
+    const collection = await renderCollection(
+      taskOptions({
+        fetcher: () => {
+          call += 1;
+          return Promise.resolve(
+            call === 1 ? projectFixtures : [{ ...projectFixtures[0], userId: 'user-2' }],
+          );
+        },
+        validate: validateProjectCollection as unknown as (value: unknown) => {
+          data: Task[];
+          workspace: string | null;
+        },
+        source: 'gateway:/api/projects',
+      }),
+    );
+    expect(collection.freshness).toBe('current');
+
+    await act(async () => {
+      await collection.revalidate();
+    });
+
+    expect(latest?.freshness).toBe('unavailable');
+    expect(latest?.data).toBeNull();
+    expect(latest?.failure).toEqual({ kind: 'invalidated', reason: 'cross-workspace' });
+  });
+
+  it('ages from current to stale and refuses mutations on stale data', async () => {
+    let fakeNow = NOW;
+    const collection = await renderCollection(
+      taskOptions({
+        clock: () => fakeNow,
+        policy: { staleAfterMs: 40 },
+        tickMs: 10,
+      }),
+    );
+    expect(collection.freshness).toBe('current');
+
+    // Age the snapshot past the policy and let the tick recompute.
+    fakeNow = NOW + 60;
+    await act(async () => {
+      await new Promise((resolve) => setTimeout(resolve, 25));
+    });
+
+    expect(latest?.freshness).toBe('stale');
+    expect(latest?.data).toEqual(taskFixtures);
+    expect(latest?.canMutate).toBe(false);
+
+    const operation = vi.fn(async () => 'result');
+    await expect(latest?.mutate(operation)).rejects.toBeInstanceOf(StaleMutationError);
+    expect(operation).not.toHaveBeenCalled();
+  });
+
+  it('recovers to current after a successful revalidation', async () => {
+    let call = 0;
+    const collection = await renderCollection(
+      taskOptions({
+        fetcher: () => {
+          call += 1;
+          return call === 1
+            ? Promise.reject(new Error('first attempt failed'))
+            : Promise.resolve(taskFixtures);
+        },
+      }),
+    );
+    expect(collection.freshness).toBe('unavailable');
+
+    await act(async () => {
+      await collection.revalidate();
+    });
+
+    expect(latest?.freshness).toBe('current');
+    expect(latest?.failure).toBeNull();
+
+    const operation = vi.fn(async (data: Task[]) => data.length);
+    await expect(latest?.mutate(operation)).resolves.toBe(taskFixtures.length);
+    expect(operation).toHaveBeenCalledOnce();
+  });
+
+  it('restores a cached snapshot as unverified stale data, then verifies it', async () => {
+    const seededVersion = seedCache('tasks');
+    const deferred = createDeferred<Task[]>();
+    const collection = await renderCollection(taskOptions({ fetcher: () => deferred.promise }));
+
+    // Restored data is situational awareness only: labeled stale, never
+    // current, and mutations are refused before verification.
+    expect(collection.freshness).toBe('stale');
+    expect(collection.data).toEqual(taskFixtures);
+    expect(collection.canMutate).toBe(false);
+    await expect(collection.mutate(vi.fn())).rejects.toBeInstanceOf(StaleMutationError);
+
+    await act(async () => {
+      deferred.resolve(taskFixtures);
+      await deferred.promise;
+    });
+
+    expect(latest?.freshness).toBe('current');
+    expect(latest?.snapshot?.version).toBe(seededVersion + 1);
+  });
+
+  it('never promotes corrupted cache data to current (cache corruption)', async () => {
+    sessionStorage.setItem('mosaic:freshness:v1:tasks', '{"data":');
+    const collection = await renderCollection(
+      taskOptions({ fetcher: () => Promise.reject(new Error('still down')) }),
+    );
+
+    expect(collection.freshness).toBe('unavailable');
+    expect(collection.data).toBeNull();
+    expect(collection.canMutate).toBe(false);
+    // The corrupted entry is dropped so it cannot come back.
+    expect(sessionStorage.getItem('mosaic:freshness:v1:tasks')).toBeNull();
+  });
+
+  it('refuses mutations while unknown or unavailable — the call itself, not just the button', async () => {
+    const deferred = createDeferred<Task[]>();
+    const unknown = await renderCollection(taskOptions({ fetcher: () => deferred.promise }));
+    const operation = vi.fn(async () => 'result');
+    await expect(unknown.mutate(operation)).rejects.toBeInstanceOf(StaleMutationError);
+    expect(operation).not.toHaveBeenCalled();
+    await act(async () => {
+      deferred.reject(new Error('failed'));
+      await deferred.promise.catch(() => undefined);
+    });
+
+    const unavailable = latest!;
+    await expect(unavailable.mutate(operation)).rejects.toBeInstanceOf(StaleMutationError);
+    expect(operation).not.toHaveBeenCalled();
+    expect(unavailable.canMutate).toBe(false);
+  });
+
+  it('degrades to stale with last-known data when a revalidation fails after success', async () => {
+    let call = 0;
+    const collection = await renderCollection(
+      taskOptions({
+        fetcher: () => {
+          call += 1;
+          return call === 1
+            ? Promise.resolve(taskFixtures)
+            : Promise.reject(new Error('connection lost'));
+        },
+      }),
+    );
+    expect(collection.freshness).toBe('current');
+
+    await act(async () => {
+      await collection.revalidate();
+    });
+
+    expect(latest?.freshness).toBe('stale');
+    expect(latest?.data).toEqual(taskFixtures);
+    const failure: FreshnessFailure | null = latest?.failure ?? null;
+    expect(failure).toEqual({ kind: 'fetch', message: 'connection lost' });
+  });
+});

apps/web/src/lib/freshness/use-fresh-collection.ts

@@ -0,0 +1,281 @@
+import { useCallback, useEffect, useMemo, useRef, useState } from 'react';
+import {
+  acceptSnapshot,
+  assertMutable,
+  computeFreshness,
+  DEFAULT_FRESHNESS_POLICY,
+  invalidationReasonLabels,
+  type FreshPayload,
+  type FreshSnapshot,
+  type FreshnessPolicy,
+  type FreshnessState,
+  type InvalidationReason,
+  StaleMutationError,
+} from './model';
+import { clearSnapshotCache, readSnapshotCache, writeSnapshotCache } from './snapshot-cache';
+
+/**
+ * Freshness-aware collection fetch hook (RI-5-001).
+ *
+ * One hook owns one gateway collection end to end: fetch, schema validation,
+ * snapshot acceptance with provenance, session-scoped last-known caching,
+ * aging, and the mutation guard. Pages consume `freshness` and never infer
+ * health from emptiness.
+ */
+
+/** Why the latest validation did not produce a current snapshot. */
+export type FreshnessFailure =
+  | { readonly kind: 'fetch'; readonly message: string }
+  | { readonly kind: 'invalidated'; readonly reason: InvalidationReason };
+
+export interface UseFreshCollectionOptions<T> {
+  /** Source identity for provenance labels, e.g. `gateway:/api/tasks`. */
+  readonly source: string;
+  /** Performs the unvalidated fetch. The hook owns abort and verification. */
+  readonly fetcher: (signal: AbortSignal) => Promise<unknown>;
+  /**
+   * Runtime schema validator. Returning `null` invalidates the payload
+   * (`schema-mismatch`) instead of letting malformed JSON flow into render.
+   */
+  readonly validate: (value: unknown) => FreshPayload<T> | null;
+  /** Overrides of the default freshness policy. */
+  readonly policy?: Partial<FreshnessPolicy>;
+  /**
+   * Session cache key for last-known snapshots. `null`/omitted disables
+   * restore. Restored snapshots are unverified: they render only as
+   * labeled `stale` data until a fetch re-verifies them.
+   */
+  readonly cacheKey?: string | null;
+  /** Injectable clock for deterministic age transitions in tests. */
+  readonly clock?: () => number;
+  /** Aging tick interval override (default derived from `staleAfterMs`). */
+  readonly tickMs?: number;
+  /** When false, no fetch runs (surfaces stay `unavailable`/`unknown`). */
+  readonly enabled?: boolean;
+}
+
+export interface FreshCollection<T> {
+  /** Last verified (or restored-unverified) snapshot, or `null`. */
+  readonly snapshot: FreshSnapshot<T> | null;
+  /** Snapshot data or `null` — never a fabricated empty collection. */
+  readonly data: T | null;
+  readonly freshness: FreshnessState;
+  /** True while a validation request is in flight. */
+  readonly validating: boolean;
+  /** Outcome of the latest failed validation, `null` when healthy. */
+  readonly failure: FreshnessFailure | null;
+  /** False unless freshness is `current`; drives disabled UI affordances. */
+  readonly canMutate: boolean;
+  /** Re-run the fetch and re-verify. Always allowed (it is a read). */
+  readonly revalidate: () => Promise<void>;
+  /**
+   * Run a state-changing operation against verified-current data only.
+   * Rejects with `StaleMutationError` on any other state — the guard fires
+   * even if a disabled button was bypassed (defense in depth).
+   */
+  readonly mutate: <R>(operation: (data: T) => Promise<R>) => Promise<R>;
+}
+
+const defaultClock = (): number => Date.now();
+
+function resolveTickMs(policy: FreshnessPolicy, override?: number): number {
+  if (override !== undefined && override > 0) return override;
+  return Math.min(5_000, Math.max(250, Math.floor(policy.staleAfterMs / 4)));
+}
+
+function isAuthFailure(caught: unknown): boolean {
+  return (
+    typeof caught === 'object' &&
+    caught !== null &&
+    'statusCode' in caught &&
+    ((caught as { statusCode?: unknown }).statusCode === 401 ||
+      (caught as { statusCode?: unknown }).statusCode === 403)
+  );
+}
+
+function fetchFailureMessage(caught: unknown): string {
+  if (caught instanceof Error && caught.message.trim().length > 0) return caught.message;
+  return 'The request failed.';
+}
+
+/** Human-readable summary of a failure for unavailable/stale notices. */
+export function describeFailure(failure: FreshnessFailure | null): string | null {
+  if (failure === null) return null;
+  if (failure.kind === 'fetch') return failure.message;
+  return `The snapshot was invalidated: ${invalidationReasonLabels[failure.reason]}.`;
+}
+
+export function useFreshCollection<T>(options: UseFreshCollectionOptions<T>): FreshCollection<T> {
+  const optionsRef = useRef(options);
+  optionsRef.current = options;
+
+  const policy = useMemo<FreshnessPolicy>(
+    () => ({ ...DEFAULT_FRESHNESS_POLICY, ...options.policy }),
+    [options.policy],
+  );
+  const policyRef = useRef(policy);
+  policyRef.current = policy;
+
+  const clockRef = useRef(options.clock ?? defaultClock);
+  clockRef.current = options.clock ?? defaultClock;
+
+  const [snapshot, setSnapshot] = useState<FreshSnapshot<T> | null>(null);
+  const [failure, setFailure] = useState<FreshnessFailure | null>(null);
+  const [unverified, setUnverified] = useState(false);
+  const [validating, setValidating] = useState(options.enabled !== false);
+  const [now, setNow] = useState(() => (options.clock ?? defaultClock)());
+
+  const snapshotRef = useRef(snapshot);
+  snapshotRef.current = snapshot;
+  const failureRef = useRef(failure);
+  failureRef.current = failure;
+  const unverifiedRef = useRef(unverified);
+  unverifiedRef.current = unverified;
+
+  const runRef = useRef(0);
+  const abortRef = useRef<AbortController | null>(null);
+
+  const revalidate = useCallback(async (): Promise<void> => {
+    const current = optionsRef.current;
+    if (current.enabled === false) {
+      setValidating(false);
+      return;
+    }
+
+    const runId = ++runRef.current;
+    abortRef.current?.abort();
+    const controller = new AbortController();
+    abortRef.current = controller;
+    setValidating(true);
+
+    let value: unknown;
+    try {
+      value = await current.fetcher(controller.signal);
+    } catch (caught) {
+      if (runRef.current !== runId || controller.signal.aborted) return;
+      if (isAuthFailure(caught)) {
+        // An unauthenticated viewer must not keep (or be served) the
+        // previous user's last-known data.
+        setSnapshot(null);
+        setUnverified(false);
+        if (current.cacheKey) clearSnapshotCache(current.cacheKey);
+      }
+      setFailure({ kind: 'fetch', message: fetchFailureMessage(caught) });
+      setValidating(false);
+      return;
+    }
+
+    if (runRef.current !== runId) return;
+
+    const result = acceptSnapshot({
+      value,
+      validate: current.validate,
+      previous: snapshotRef.current,
+      policy: policyRef.current,
+      source: current.source,
+      now: clockRef.current(),
+    });
+
+    if (result.outcome === 'accepted') {
+      setSnapshot(result.snapshot);
+      setUnverified(false);
+      setFailure(null);
+      if (current.cacheKey) writeSnapshotCache(current.cacheKey, result.snapshot);
+    } else {
+      if (result.reason === 'cross-workspace') {
+        // Data verified for a different workspace must not linger as
+        // last-known situational awareness either.
+        setSnapshot(null);
+        setUnverified(false);
+      }
+      if (current.cacheKey) clearSnapshotCache(current.cacheKey);
+      setFailure({ kind: 'invalidated', reason: result.reason });
+    }
+    setValidating(false);
+  }, []);
+
+  // Restore the last-known snapshot (unverified) and run the first fetch.
+  useEffect(() => {
+    if (optionsRef.current.enabled === false) {
+      setValidating(false);
+      return;
+    }
+
+    const cacheKey = optionsRef.current.cacheKey;
+    if (cacheKey) {
+      const restored = readSnapshotCache<T>({
+        key: cacheKey,
+        workspace: policyRef.current.workspace,
+        policy: policyRef.current,
+        validate: optionsRef.current.validate,
+      });
+      if (restored.outcome === 'hit') {
+        setSnapshot(restored.snapshot);
+        setUnverified(true);
+      } else if (restored.outcome === 'invalidated') {
+        // A corrupted/foreign/regressed entry is dropped immediately; it must
+        // never surface as data. The fetch decides the visible state.
+        clearSnapshotCache(cacheKey);
+      }
+    }
+
+    void revalidate();
+
+    return () => {
+      abortRef.current?.abort();
+    };
+    // Mount-once by design: `revalidate` is stable and reads live options
+    // through refs, so it never needs to re-run when options change.
+    // Route-param pages remount this hook via an identity `key` instead.
+  }, [revalidate]);
+
+  // Aging tick: recomputes freshness as the snapshot ages past the policy.
+  useEffect(() => {
+    const interval = setInterval(
+      () => {
+        setNow(clockRef.current());
+      },
+      resolveTickMs(policyRef.current, optionsRef.current.tickMs),
+    );
+    return () => clearInterval(interval);
+  }, []);
+
+  const freshness = useMemo<FreshnessState>(() => {
+    if (snapshot === null) return validating ? 'unknown' : 'unavailable';
+    return computeFreshness({
+      snapshot,
+      policy,
+      now,
+      degraded: failure !== null || unverified,
+    });
+    // `now` from state covers age; refs inside computeFreshness are pure.
+  }, [snapshot, validating, failure, unverified, now, policy]);
+
+  const canMutate = freshness === 'current';
+
+  const mutate = useCallback(async <R>(operation: (data: T) => Promise<R>): Promise<R> => {
+    const currentSnapshot = snapshotRef.current;
+    // No verified snapshot at all: with nothing verified there is nothing
+    // current to mutate, regardless of the recorded failure.
+    if (currentSnapshot === null) throw new StaleMutationError('unavailable');
+    const state = computeFreshness({
+      snapshot: currentSnapshot,
+      policy: policyRef.current,
+      now: clockRef.current(),
+      degraded: failureRef.current !== null || unverifiedRef.current,
+    });
+    assertMutable(state);
+    return operation(currentSnapshot.data);
+  }, []);
+
+  return {
+    snapshot,
+    data: snapshot === null ? null : snapshot.data,
+    freshness,
+    validating,
+    failure,
+    canMutate,
+    revalidate,
+    mutate,
+  };
+}

apps/web/src/lib/freshness/validators.spec.ts

@@ -0,0 +1,103 @@
+import { describe, expect, it } from 'vitest';
+import type { Mission, Project, Task } from '@/lib/types';
+import {
+  validateMissionCollection,
+  validateProjectCollection,
+  validateProjectEntity,
+  validateTaskCollection,
+} from './validators';
+import { missionFixtures, projectFixtures, taskFixtures } from '@/spa/pages/page-fixtures';
+
+describe('validateTaskCollection', () => {
+  it('accepts a well-formed task collection', () => {
+    expect(validateTaskCollection(taskFixtures)).toEqual({
+      data: taskFixtures,
+      workspace: null,
+    });
+  });
+
+  it('accepts an empty collection (a healthy empty state is a valid payload)', () => {
+    expect(validateTaskCollection([])).toEqual({ data: [], workspace: null });
+  });
+
+  it.each([
+    ['not an array', { items: [] }],
+    ['item is not an object', ['nope']],
+    ['missing id', [{ ...(taskFixtures[0] as Task), id: undefined }]],
+    ['missing title', [{ ...(taskFixtures[0] as Task), title: undefined }]],
+    ['unknown status enum', [{ ...(taskFixtures[0] as Task), status: 'finished' }]],
+    ['unknown priority enum', [{ ...(taskFixtures[0] as Task), priority: 'urgent' }]],
+    ['tags of the wrong type', [{ ...(taskFixtures[0] as Task), tags: 'spa' }]],
+    ['metadata of the wrong type', [{ ...(taskFixtures[0] as Task), metadata: 'notes' }]],
+    ['createdAt of the wrong type', [{ ...(taskFixtures[0] as Task), createdAt: 1234 }]],
+    ['null sneaks past a required string', [{ ...(taskFixtures[0] as Task), title: null }]],
+  ])('rejects a malformed payload: %s', (_label, value) => {
+    expect(validateTaskCollection(value)).toBeNull();
+  });
+});
+
+describe('validateMissionCollection', () => {
+  it('accepts a well-formed mission collection', () => {
+    expect(validateMissionCollection(missionFixtures)).toEqual({
+      data: missionFixtures,
+      workspace: null,
+    });
+  });
+
+  it.each([
+    ['not an array', null],
+    ['item missing name', [{ ...(missionFixtures[0] as Mission), name: 42 }]],
+    ['unknown status enum', [{ ...(missionFixtures[0] as Mission), status: 'canceled' }]],
+    ['projectId of the wrong type', [{ ...(missionFixtures[0] as Mission), projectId: 7 }]],
+  ])('rejects a malformed payload: %s', (_label, value) => {
+    expect(validateMissionCollection(value)).toBeNull();
+  });
+});
+
+describe('validateProjectCollection', () => {
+  it('accepts a uniform workspace-scoped collection and reports its workspace', () => {
+    expect(validateProjectCollection(projectFixtures)).toEqual({
+      data: projectFixtures,
+      workspace: 'user-1',
+    });
+  });
+
+  it('accepts an empty collection with no workspace identity', () => {
+    expect(validateProjectCollection([])).toEqual({ data: [], workspace: null });
+  });
+
+  it.each([
+    ['not an array', 42],
+    ['item missing userId', [{ ...(projectFixtures[0] as Project), userId: undefined }]],
+    ['unknown status enum', [{ ...(projectFixtures[0] as Project), status: 'live' }]],
+    ['description of the wrong type', [{ ...(projectFixtures[0] as Project), description: 1 }]],
+  ])('rejects a malformed payload: %s', (_label, value) => {
+    expect(validateProjectCollection(value)).toBeNull();
+  });
+
+  it('rejects a collection mixing workspace identities (cross-workspace leak)', () => {
+    const mixed = [
+      projectFixtures[0] as Project,
+      { ...(projectFixtures[1] as Project), userId: 'user-2' },
+    ];
+    expect(validateProjectCollection(mixed)).toBeNull();
+  });
+});
+
+describe('validateProjectEntity', () => {
+  it('accepts a well-formed project and reports its workspace', () => {
+    expect(validateProjectEntity(projectFixtures[0])).toEqual({
+      data: projectFixtures[0],
+      workspace: 'user-1',
+    });
+  });
+
+  it.each([
+    ['not an object', 'project-1'],
+    ['null', null],
+    ['array', [projectFixtures[0]]],
+    ['missing userId', [{ ...(projectFixtures[0] as Project), userId: null }]],
+  ])('rejects a malformed entity: %s', (_label, value) => {
+    expect(validateProjectEntity(value)).toBeNull();
+  });
+});

apps/web/src/lib/freshness/validators.ts

@@ -0,0 +1,135 @@
+import type { Mission, Project, Task, MissionStatus, TaskPriority, TaskStatus } from '@/lib/types';
+import type { FreshPayload } from './model';
+
+/**
+ * Runtime schema validators for gateway collections (RI-5-001).
+ *
+ * `api<T>()` returns untrusted JSON cast to `T`; these validators are the
+ * seam where a malformed response becomes an explicit schema mismatch
+ * instead of flowing into the render path as if it were healthy data.
+ */
+
+const taskStatuses: readonly TaskStatus[] = [
+  'not-started',
+  'in-progress',
+  'blocked',
+  'done',
+  'cancelled',
+];
+const taskPriorities: readonly TaskPriority[] = ['critical', 'high', 'medium', 'low'];
+const missionStatuses: readonly MissionStatus[] = [
+  'planning',
+  'active',
+  'paused',
+  'completed',
+  'failed',
+];
+const projectStatuses: readonly Project['status'][] = ['active', 'paused', 'completed', 'archived'];
+
+function isRecord(value: unknown): value is Record<string, unknown> {
+  return typeof value === 'object' && value !== null && !Array.isArray(value);
+}
+
+function isString(value: unknown): value is string {
+  return typeof value === 'string';
+}
+
+function isNullableString(value: unknown): value is string | null {
+  return value === null || typeof value === 'string';
+}
+
+function isOneOf<T extends string>(value: unknown, allowed: readonly T[]): value is T {
+  return typeof value === 'string' && (allowed as readonly string[]).includes(value);
+}
+
+function isNullableRecord(value: unknown): value is Record<string, unknown> | null {
+  return value === null || isRecord(value);
+}
+
+function isNullableStringArray(value: unknown): value is string[] | null {
+  if (value === null) return true;
+  if (!Array.isArray(value)) return false;
+  return value.every((item) => typeof item === 'string');
+}
+
+function isIsoLike(value: unknown): value is string {
+  return typeof value === 'string' && value.length > 0;
+}
+
+function isTask(value: unknown): value is Task {
+  if (!isRecord(value)) return false;
+  return (
+    isString(value['id']) &&
+    isString(value['title']) &&
+    isOneOf(value['status'], taskStatuses) &&
+    isOneOf(value['priority'], taskPriorities) &&
+    isNullableString(value['projectId']) &&
+    isNullableString(value['missionId']) &&
+    isNullableString(value['assignee']) &&
+    isNullableStringArray(value['tags']) &&
+    isNullableRecord(value['metadata']) &&
+    isNullableString(value['dueDate']) &&
+    isIsoLike(value['createdAt']) &&
+    isIsoLike(value['updatedAt'])
+  );
+}
+
+/** Tasks carry no workspace identity; scope falls back to the policy. */
+export function validateTaskCollection(value: unknown): FreshPayload<Task[]> | null {
+  if (!Array.isArray(value) || !value.every(isTask)) return null;
+  return { data: value as Task[], workspace: null };
+}
+
+function isMission(value: unknown): value is Mission {
+  if (!isRecord(value)) return false;
+  return (
+    isString(value['id']) &&
+    isString(value['name']) &&
+    isOneOf(value['status'], missionStatuses) &&
+    isNullableString(value['projectId']) &&
+    isNullableString(value['description']) &&
+    isNullableRecord(value['metadata']) &&
+    isIsoLike(value['createdAt']) &&
+    isIsoLike(value['updatedAt'])
+  );
+}
+
+/** Missions carry no workspace identity; scope falls back to the policy. */
+export function validateMissionCollection(value: unknown): FreshPayload<Mission[]> | null {
+  if (!Array.isArray(value) || !value.every(isMission)) return null;
+  return { data: value as Mission[], workspace: null };
+}
+
+function isProject(value: unknown): value is Project {
+  if (!isRecord(value)) return false;
+  return (
+    isString(value['id']) &&
+    isString(value['name']) &&
+    isOneOf(value['status'], projectStatuses) &&
+    isString(value['userId']) &&
+    isNullableString(value['description']) &&
+    isNullableRecord(value['metadata']) &&
+    isIsoLike(value['createdAt']) &&
+    isIsoLike(value['updatedAt'])
+  );
+}
+
+/**
+ * Projects are workspace-scoped: every item must carry the same `userId`.
+ * A collection mixing identities (cross-workspace leak) is a schema
+ * mismatch; the uniform `userId` becomes the snapshot workspace.
+ */
+export function validateProjectCollection(value: unknown): FreshPayload<Project[]> | null {
+  if (!Array.isArray(value) || !value.every(isProject)) return null;
+  const projects = value as Project[];
+  const workspaces = new Set(projects.map((project) => project.userId));
+  if (workspaces.size > 1) return null;
+  return { data: projects, workspace: projects.length > 0 ? projects[0]!.userId : null };
+}
+
+/** Single project entity (project detail primary collection). */
+export function validateProjectEntity(value: unknown): FreshPayload<Project> | null {
+  if (!isProject(value)) return null;
+  const project = value as Project;
+  return { data: project, workspace: project.userId };
+}

apps/web/src/spa/pages/project-detail.spec.tsx

@@ -35,6 +35,7 @@ afterEach(async () => {
   document.body.replaceChildren();
   root = null;
   apiMock.mockReset();
+  sessionStorage.clear();
 });
 
 async function renderProjectDetailPage(): Promise<ReturnType<typeof createMemoryRouter>> {
@@ -64,21 +65,49 @@ function clickButtonByText(text: string): void {
   button.dispatchEvent(new MouseEvent('click', { bubbles: true }));
 }
 
+async function flushAct(): Promise<void> {
+  await act(async () => {
+    await Promise.resolve();
+  });
+}
+
+interface Deferred<T> {
+  promise: Promise<T>;
+  resolve: (value: T) => void;
+}
+
+function createDeferred<T>(): Deferred<T> {
+  let resolve!: (value: T) => void;
+  const promise = new Promise<T>((res) => {
+    resolve = res;
+  });
+  return { promise, resolve };
+}
+
+const projectOneTasks = taskFixtures.filter((task) => task.projectId === 'project-1');
+
+function mockHealthyLoad(): void {
+  apiMock
+    .mockResolvedValueOnce(projectFixtures[0])
+    .mockResolvedValueOnce(missionFixtures)
+    .mockResolvedValueOnce(projectOneTasks);
+}
+
 describe('ProjectDetailPage', () => {
   it('loads the project, tasks, missions, and optional PRD content for the active project', async () => {
-    apiMock
+    mockHealthyLoad();
-      .mockResolvedValueOnce(projectFixtures[0])
-      .mockResolvedValueOnce(missionFixtures)
-      .mockResolvedValueOnce(taskFixtures.filter((task) => task.projectId === 'project-1'));
 
     await renderProjectDetailPage();
 
-    expect(apiMock.mock.calls).toEqual([
+    expect(apiMock.mock.calls.map((call) => call[0])).toEqual([
-      ['/api/projects/project-1'],
+      '/api/projects/project-1',
-      ['/api/missions'],
+      '/api/missions',
-      ['/api/tasks?projectId=project-1'],
+      '/api/tasks?projectId=project-1',
     ]);
 
+    expect(container.querySelector('[data-freshness]')?.getAttribute('data-freshness')).toBe(
+      'current',
+    );
     expect(container.textContent).toContain('Mosaic Stack');
     expect(container.textContent).toContain('Route /projects/:id');
     expect(container.textContent).toContain('Tasks');
@@ -101,10 +130,7 @@ describe('ProjectDetailPage', () => {
   });
 
   it('opens and closes the existing read-only task modal from the tasks tab', async () => {
-    apiMock
+    mockHealthyLoad();
-      .mockResolvedValueOnce(projectFixtures[0])
-      .mockResolvedValueOnce(missionFixtures)
-      .mockResolvedValueOnce(taskFixtures.filter((task) => task.projectId === 'project-1'));
 
     await renderProjectDetailPage();
 
@@ -134,35 +160,153 @@ describe('ProjectDetailPage', () => {
     expect(container.querySelector('[role="dialog"]')).toBeNull();
   });
 
-  it('renders the project with an empty missions tab when the missions request fails', async () => {
+  it('shows verified completion verdicts when the task collection is current', async () => {
-    apiMock
+    mockHealthyLoad();
-      .mockResolvedValueOnce(projectFixtures[0])
-      .mockRejectedValueOnce(new Error('Missions request failed'))
-      .mockResolvedValueOnce(taskFixtures.filter((task) => task.projectId === 'project-1'));
 
     await renderProjectDetailPage();
 
-    expect(container.textContent).toContain('Mosaic Stack');
+    const doneCard = [...container.querySelectorAll('div')].find(
-    expect(container.querySelector('[role="alert"]')).toBeNull();
+      (candidate) => candidate.textContent === 'Done1',
-
+    );
-    await act(async () => {
+    expect(doneCard).toBeTruthy();
-      clickButtonByText('Missions (0)');
+    const inProgressCard = [...container.querySelectorAll('div')].find(
-    });
+      (candidate) => candidate.textContent === 'In Progress1',
-
+    );
-    expect(container.textContent).toContain('No missions for this project');
+    expect(inProgressCard).toBeTruthy();
   });
 
-  it('renders a visible alert when the project request fails and lets the user navigate back', async () => {
+  it('renders an explicit unavailable missions tab when the missions request fails (partial, not empty)', async () => {
+    apiMock
+      .mockResolvedValueOnce(projectFixtures[0])
+      .mockRejectedValueOnce(new Error('Missions request failed'))
+      .mockResolvedValueOnce(projectOneTasks);
+
+    await renderProjectDetailPage();
+
+    // Secondary failure degrades the surface to partial; the project itself
+    // still renders.
+    expect(container.querySelector('[data-freshness]')?.getAttribute('data-freshness')).toBe(
+      'partial',
+    );
+    expect(container.textContent).toContain('Mosaic Stack');
+    const partial = container.querySelector('[role="status"]');
+    expect(partial?.textContent).toContain('Missions');
+    expect(partial?.textContent).toContain('unavailable');
+
+    await act(async () => {
+      clickButtonByText('Missions (?)');
+    });
+
+    const alert = container.querySelector('[role="alert"]');
+    expect(alert?.textContent).toContain('Missions request failed');
+    // Negative control: a failed fetch must not look like an empty list.
+    expect(container.textContent).not.toContain('No missions for this project');
+  });
+
+  it('marks derived verdicts unknown when the tasks collection is unavailable', async () => {
+    apiMock
+      .mockResolvedValueOnce(projectFixtures[0])
+      .mockResolvedValueOnce(missionFixtures)
+      .mockRejectedValueOnce(new Error('Tasks request failed'));
+
+    await renderProjectDetailPage();
+
+    expect(container.querySelector('[data-freshness]')?.getAttribute('data-freshness')).toBe(
+      'partial',
+    );
+
+    // Completion verdicts become unknown ('?') — never green counts.
+    for (const label of ['Done', 'In Progress', 'Blocked', 'Tasks']) {
+      const unknownCard = [...container.querySelectorAll('div')].find(
+        (candidate) => candidate.textContent === `${label}?`,
+      );
+      expect(unknownCard, `expected ${label} card to render ?`).toBeTruthy();
+    }
+    // Negative control: no green "Done 1" verdict anywhere.
+    expect(
+      [...container.querySelectorAll('div')].some((candidate) => candidate.textContent === 'Done1'),
+    ).toBe(false);
+
+    await act(async () => {
+      clickButtonByText('Tasks (?)');
+    });
+
+    const alert = container.querySelector('[role="alert"]');
+    expect(alert?.textContent).toContain('Tasks request failed');
+    // Negative control: no healthy empty task list from a failed fetch.
+    expect(container.textContent).not.toContain('No tasks found');
+    expect(container.querySelector('table')).toBeNull();
+  });
+
+  it('recovers a partial surface to current after revalidation', async () => {
+    apiMock
+      .mockResolvedValueOnce(projectFixtures[0])
+      .mockResolvedValueOnce(missionFixtures)
+      .mockRejectedValueOnce(new Error('Tasks request failed'))
+      .mockResolvedValueOnce(projectFixtures[0])
+      .mockResolvedValueOnce(missionFixtures)
+      .mockResolvedValueOnce(projectOneTasks);
+
+    await renderProjectDetailPage();
+    expect(container.querySelector('[data-freshness]')?.getAttribute('data-freshness')).toBe(
+      'partial',
+    );
+
+    await act(async () => {
+      clickButtonByText('Revalidate');
+    });
+    await flushAct();
+
+    expect(container.querySelector('[data-freshness]')?.getAttribute('data-freshness')).toBe(
+      'current',
+    );
+    expect(
+      [...container.querySelectorAll('div')].some((candidate) => candidate.textContent === 'Done1'),
+    ).toBe(true);
+  });
+
+  it("never shows one project's data on another project's route after navigation", async () => {
+    mockHealthyLoad();
+
+    const router = await renderProjectDetailPage();
+    expect(container.textContent).toContain('Mosaic Stack');
+
+    const deferred = createDeferred<(typeof projectFixtures)[number]>();
+    apiMock
+      .mockResolvedValueOnce(deferred.promise)
+      .mockResolvedValueOnce([])
+      .mockResolvedValueOnce([]);
+
+    await act(async () => {
+      await router.navigate('/projects/project-2');
+    });
+
+    // While project-2 loads, nothing from project-1 may render on its route.
+    expect(container.textContent).toContain('Loading project...');
+    expect(container.textContent).not.toContain('Mosaic Stack');
+    expect(container.textContent).not.toContain('Route /projects/:id');
+
+    await act(async () => {
+      deferred.resolve(projectFixtures[1]!);
+      await deferred.promise;
+    });
+
+    expect(container.textContent).toContain('Agent Runtime');
+    expect(apiMock.mock.calls[3]?.[0]).toBe('/api/projects/project-2');
+  });
+
+  it('renders a visible unavailable state when the project request fails and lets the user navigate back', async () => {
     apiMock
       .mockRejectedValueOnce(new Error('Project request failed'))
       .mockResolvedValueOnce(missionFixtures)
-      .mockResolvedValueOnce(taskFixtures.filter((task) => task.projectId === 'project-1'));
+      .mockResolvedValueOnce(projectOneTasks);
 
     const router = await renderProjectDetailPage();
 
     const alert = container.querySelector('[role="alert"]');
     expect(alert).toBeTruthy();
     expect(alert?.textContent).toContain('Project request failed');
+    expect(alert?.textContent).toContain('not an empty result');
     expect(container.textContent).not.toContain('Mosaic Stack');
 
     await act(async () => {

apps/web/src/spa/pages/project-detail.tsx

@@ -1,14 +1,30 @@
-import { useEffect, useState, type ReactElement } from 'react';
+import { useState, type ReactElement } from 'react';
 import { useNavigate, useParams } from 'react-router-dom';
 import { MissionTimeline } from '@/components/projects/mission-timeline';
 import { PrdViewer } from '@/components/projects/prd-viewer';
 import { TaskDetailModal } from '@/components/tasks/task-detail-modal';
 import { TaskListView } from '@/components/tasks/task-list-view';
 import { TaskStatusSummary } from '@/components/tasks/task-status-summary';
+import {
+  PartialDataNotice,
+  StaleDataNotice,
+  UnavailableDataNotice,
+} from '@/components/freshness/freshness-notices';
 import { api } from '@/lib/api';
 import { cn } from '@/lib/cn';
 import type { Mission, Project, Task, TaskStatus } from '@/lib/types';
-import { getErrorMessage } from './page-errors';
+import {
+  combineFreshness,
+  UNKNOWN_VERDICT,
+  verdictValue,
+  type FreshSnapshot,
+} from '@/lib/freshness/model';
+import { describeFailure, useFreshCollection } from '@/lib/freshness/use-fresh-collection';
+import {
+  validateMissionCollection,
+  validateProjectEntity,
+  validateTaskCollection,
+} from '@/lib/freshness/validators';
 
 type Tab = 'overview' | 'tasks' | 'missions' | 'prd';
 
@@ -51,73 +67,62 @@ function TabButton({ id, label, activeTab, onClick }: TabButtonProps): ReactElem
   );
 }
 
+/** Remounts per project id so no state from one project renders for another. */
 export function ProjectDetailPage(): ReactElement {
   const { id = '' } = useParams();
+  return <ProjectDetail id={id} key={id} />;
+}
+
+function ProjectDetail({ id }: { id: string }): ReactElement {
   const navigate = useNavigate();
-  const [project, setProject] = useState<Project | null>(null);
+  const enabled = id.length > 0;
-  const [missions, setMissions] = useState<Mission[]>([]);
+
-  const [tasks, setTasks] = useState<Task[]>([]);
+  // Primary collection gates the surface; missions and tasks are secondaries
-  const [loading, setLoading] = useState(true);
+  // whose failures degrade the surface to `partial` instead of rendering
-  const [error, setError] = useState<string | null>(null);
+  // empty healthy lists.
+  const project = useFreshCollection<Project>({
+    source: `gateway:/api/projects/${id}`,
+    fetcher: (signal) => api<unknown>(`/api/projects/${id}`, { signal }),
+    validate: validateProjectEntity,
+    // No last-known restore: the entity carries workspace identity that
+    // cannot be scope-checked before display (see ProjectsPage note).
+    enabled,
+  });
+  const missions = useFreshCollection<Mission[]>({
+    source: 'gateway:/api/missions',
+    fetcher: (signal) => api<unknown>('/api/missions', { signal }),
+    validate: validateMissionCollection,
+    cacheKey: enabled ? 'missions' : null,
+    enabled,
+  });
+  const tasks = useFreshCollection<Task[]>({
+    source: `gateway:/api/tasks?projectId=${id}`,
+    fetcher: (signal) => api<unknown>(`/api/tasks?projectId=${id}`, { signal }),
+    validate: validateTaskCollection,
+    cacheKey: enabled ? `project-tasks:${id}` : null,
+    enabled,
+  });
+
   const [activeTab, setActiveTab] = useState<Tab>('overview');
   const [taskFilter, setTaskFilter] = useState<TaskStatus | 'all'>('all');
   const [selectedTask, setSelectedTask] = useState<Task | null>(null);
 
-  useEffect(() => {
+  const surface = combineFreshness(project.freshness, [missions.freshness, tasks.freshness]);
-    if (!id) {
+  const tasksVerified = tasks.freshness === 'current';
-      setError('Project id is missing.');
+  const projectMissions = missions.data?.filter((mission) => mission.projectId === id) ?? null;
-      setLoading(false);
-      return;
-    }
 
-    let cancelled = false;
+  const retryAll = (): void => {
-    setLoading(true);
+    void Promise.all([project.revalidate(), missions.revalidate(), tasks.revalidate()]);
-    setError(null);
+  };
 
-    void Promise.all([
+  if (!enabled) {
-      api<Project>('/api/projects/' + id),
-      api<Mission[]>('/api/missions').catch(() => [] as Mission[]),
-      api<Task[]>('/api/tasks?projectId=' + id).catch(() => [] as Task[]),
-    ])
-      .then(([loadedProject, allMissions, loadedTasks]) => {
-        if (cancelled) return;
-        setProject(loadedProject);
-        setMissions(allMissions.filter((mission) => mission.projectId === id));
-        setTasks(loadedTasks);
-      })
-      .catch((caught: unknown) => {
-        if (cancelled) return;
-        setError(getErrorMessage(caught, 'Failed to load project.'));
-      })
-      .finally(() => {
-        if (cancelled) return;
-        setLoading(false);
-      });
-
-    return () => {
-      cancelled = true;
-    };
-  }, [id]);
-
-  if (loading) {
-    return (
-      <div className="flex min-h-screen flex-col px-4 py-6 sm:px-6">
-        <header className="mb-6 border-b px-1 pb-3">
-          <h1 className="text-2xl font-semibold">Project</h1>
-        </header>
-        <p className="py-16 text-center text-sm text-text-muted">Loading project...</p>
-      </div>
-    );
-  }
-
-  if (error || !project) {
     return (
       <div className="flex min-h-screen flex-col px-4 py-6 sm:px-6">
         <header className="mb-6 border-b px-1 pb-3">
           <h1 className="text-2xl font-semibold">Project</h1>
         </header>
         <div role="alert" className="rounded-lg border border-error/40 px-4 py-3 text-sm">
-          {error ?? 'Project not found.'}
+          Project id is missing.
         </div>
         <button
           type="button"
@@ -130,18 +135,81 @@ export function ProjectDetailPage(): ReactElement {
     );
   }
 
+  if (project.freshness === 'unknown') {
+    return (
+      <div className="flex min-h-screen flex-col px-4 py-6 sm:px-6">
+        <header className="mb-6 border-b px-1 pb-3">
+          <h1 className="text-2xl font-semibold">Project</h1>
+        </header>
+        <p className="py-16 text-center text-sm text-text-muted">Loading project...</p>
+      </div>
+    );
+  }
+
+  if (project.freshness === 'unavailable' || project.data === null) {
+    return (
+      <div className="flex min-h-screen flex-col px-4 py-6 sm:px-6">
+        <header className="mb-6 border-b px-1 pb-3">
+          <h1 className="text-2xl font-semibold">Project</h1>
+        </header>
+        <UnavailableDataNotice
+          title="This project"
+          detail={describeFailure(project.failure)}
+          onRetry={retryAll}
+        />
+        <button
+          type="button"
+          onClick={() => navigate('/projects')}
+          className="mt-4 w-fit text-sm underline"
+        >
+          Back to projects
+        </button>
+      </div>
+    );
+  }
+
+  const projectTasks = tasks.data ?? null;
   const filteredTasks =
-    taskFilter === 'all' ? tasks : tasks.filter((task) => task.status === taskFilter);
+    projectTasks === null
-  const prdContent = getPrdContent(project);
+      ? []
+      : taskFilter === 'all'
+        ? projectTasks
+        : projectTasks.filter((task) => task.status === taskFilter);
+
+  // Derived completion verdicts: unknown (never green) unless the task
+  // collection is verified current.
+  const doneCount = projectTasks?.filter((task) => task.status === 'done').length ?? 0;
+  const inProgressCount = projectTasks?.filter((task) => task.status === 'in-progress').length ?? 0;
+  const blockedCount = projectTasks?.filter((task) => task.status === 'blocked').length ?? 0;
+
+  const prdContent = getPrdContent(project.data);
   const tabs: Array<{ id: Tab; label: string }> = [
     { id: 'overview', label: 'Overview' },
-    { id: 'tasks', label: `Tasks (${tasks.length})` },
+    {
-    { id: 'missions', label: `Missions (${missions.length})` },
+      id: 'tasks',
+      label: `Tasks (${projectTasks === null ? UNKNOWN_VERDICT : projectTasks.length})`,
+    },
+    {
+      id: 'missions',
+      label: `Missions (${projectMissions === null ? UNKNOWN_VERDICT : projectMissions.length})`,
+    },
     ...(prdContent ? [{ id: 'prd' as const, label: 'PRD' }] : []),
   ];
 
+  const staleSnapshot: FreshSnapshot<unknown> | null =
+    project.freshness === 'stale'
+      ? project.snapshot
+      : missions.freshness === 'stale'
+        ? missions.snapshot
+        : tasks.freshness === 'stale'
+          ? tasks.snapshot
+          : null;
+  const missingSections: string[] = [];
+  if (missions.freshness === 'unavailable') missingSections.push('Missions');
+  if (tasks.freshness === 'unavailable') missingSections.push('Tasks');
+
   return (
-    <div className="flex min-h-screen flex-col px-4 py-6 sm:px-6">
+    <div data-freshness={surface} className="flex min-h-screen flex-col px-4 py-6 sm:px-6">
       <header className="mb-6 border-b px-1 pb-3">
         <nav className="mb-4 flex items-center gap-2 text-sm text-text-muted">
           <button
@@ -152,49 +220,64 @@ export function ProjectDetailPage(): ReactElement {
             Projects
           </button>
           <span>/</span>
-          <span className="text-text-primary">{project.name}</span>
+          <span className="text-text-primary">{project.data.name}</span>
         </nav>
 
         <div className="flex items-start justify-between gap-4">
           <div>
             <div className="flex items-center gap-3">
-              <h1 className="text-2xl font-semibold text-text-primary">{project.name}</h1>
+              <h1 className="text-2xl font-semibold text-text-primary">{project.data.name}</h1>
               <span
                 className={cn(
                   'rounded-full px-2 py-0.5 text-xs',
-                  projectStatusColors[project.status] ?? 'bg-gray-600/20 text-gray-400',
+                  projectStatusColors[project.data.status] ?? 'bg-gray-600/20 text-gray-400',
                 )}
               >
-                {project.status}
+                {project.data.status}
               </span>
             </div>
-            {project.description ? (
+            {project.data.description ? (
-              <p className="mt-1 text-sm text-text-muted">{project.description}</p>
+              <p className="mt-1 text-sm text-text-muted">{project.data.description}</p>
             ) : null}
             <p className="mt-2 text-xs text-text-muted">
-              Created {new Date(project.createdAt).toLocaleDateString()} · Updated{' '}
+              Created {new Date(project.data.createdAt).toLocaleDateString()} · Updated{' '}
-              {new Date(project.updatedAt).toLocaleDateString()}
+              {new Date(project.data.updatedAt).toLocaleDateString()}
             </p>
           </div>
         </div>
       </header>
 
+      {staleSnapshot !== null ? (
+        <div className="mb-6">
+          <StaleDataNotice label={staleSnapshot} onRetry={retryAll} />
+        </div>
+      ) : null}
+
+      {missingSections.length > 0 ? (
+        <div className="mb-6">
+          <PartialDataNotice missing={missingSections} onRetry={retryAll} />
+        </div>
+      ) : null}
+
       <div className="mb-6 grid grid-cols-2 gap-3 sm:grid-cols-4">
-        <StatCard label="Tasks" value={String(tasks.length)} />
+        <StatCard
+          label="Tasks"
+          value={projectTasks === null ? UNKNOWN_VERDICT : String(projectTasks.length)}
+        />
         <StatCard
           label="Done"
-          value={String(tasks.filter((task) => task.status === 'done').length)}
+          value={verdictValue(tasksVerified, String(doneCount))}
-          valueClass="text-success"
+          valueClass={tasksVerified ? 'text-success' : undefined}
         />
         <StatCard
           label="In Progress"
-          value={String(tasks.filter((task) => task.status === 'in-progress').length)}
+          value={verdictValue(tasksVerified, String(inProgressCount))}
-          valueClass="text-blue-400"
+          valueClass={tasksVerified ? 'text-blue-400' : undefined}
         />
         <StatCard
           label="Blocked"
-          value={String(tasks.filter((task) => task.status === 'blocked').length)}
+          value={verdictValue(tasksVerified, String(blockedCount))}
-          valueClass={tasks.some((task) => task.status === 'blocked') ? 'text-error' : undefined}
+          valueClass={tasksVerified && blockedCount > 0 ? 'text-error' : undefined}
         />
       </div>
 
@@ -211,23 +294,43 @@ export function ProjectDetailPage(): ReactElement {
       </div>
 
       {activeTab === 'overview' ? (
-        <OverviewTab project={project} missions={missions} tasks={tasks} />
+        <OverviewTab project={project.data} missions={projectMissions} tasks={projectTasks} />
       ) : null}
 
       {activeTab === 'tasks' ? (
         <div>
-          <div className="mb-4">
+          {projectTasks === null ? (
-            <TaskStatusSummary
+            <UnavailableDataNotice
-              tasks={tasks}
+              title="Tasks"
-              activeFilter={taskFilter}
+              detail={describeFailure(tasks.failure)}
-              onFilterChange={setTaskFilter}
+              onRetry={retryAll}
             />
-          </div>
+          ) : (
-          <TaskListView tasks={filteredTasks} onTaskClick={setSelectedTask} />
+            <>
+              <div className="mb-4">
+                <TaskStatusSummary
+                  tasks={projectTasks}
+                  activeFilter={taskFilter}
+                  onFilterChange={setTaskFilter}
+                />
+              </div>
+              <TaskListView tasks={filteredTasks} onTaskClick={setSelectedTask} />
+            </>
+          )}
         </div>
       ) : null}
 
-      {activeTab === 'missions' ? <MissionTimeline missions={missions} /> : null}
+      {activeTab === 'missions' ? (
+        projectMissions === null ? (
+          <UnavailableDataNotice
+            title="Missions"
+            detail={describeFailure(missions.failure)}
+            onRetry={retryAll}
+          />
+        ) : (
+          <MissionTimeline missions={projectMissions} />
+        )
+      ) : null}
 
       {activeTab === 'prd' && prdContent ? (
         <div className="rounded-lg border border-surface-border bg-surface-card p-6">
@@ -248,18 +351,26 @@ function OverviewTab({
   tasks,
 }: {
   project: Project;
-  missions: Mission[];
+  missions: Mission[] | null;
-  tasks: Task[];
+  tasks: Task[] | null;
 }): ReactElement {
-  const recentTasks = [...tasks]
+  const recentTasks =
-    .sort((left, right) => new Date(right.updatedAt).getTime() - new Date(left.updatedAt).getTime())
+    tasks === null
-    .slice(0, 5);
+      ? null
+      : [...tasks]
+          .sort(
+            (left, right) =>
+              new Date(right.updatedAt).getTime() - new Date(left.updatedAt).getTime(),
+          )
+          .slice(0, 5);
 
   return (
     <div className="grid gap-6 lg:grid-cols-2">
       <section>
         <h2 className="mb-3 text-sm font-semibold text-text-secondary">Recent Tasks</h2>
-        {recentTasks.length === 0 ? (
+        {recentTasks === null ? (
+          <UnavailableDataNotice title="Tasks" />
+        ) : recentTasks.length === 0 ? (
           <div className="rounded-lg border border-surface-border bg-surface-card p-4 text-center">
             <p className="text-sm text-text-muted">No tasks yet</p>
           </div>
@@ -287,7 +398,9 @@ function OverviewTab({
 
       <section>
         <h2 className="mb-3 text-sm font-semibold text-text-secondary">Missions</h2>
-        {missions.length === 0 ? (
+        {missions === null ? (
+          <UnavailableDataNotice title="Missions" />
+        ) : missions.length === 0 ? (
           <div className="rounded-lg border border-surface-border bg-surface-card p-4 text-center">
             <p className="text-sm text-text-muted">No missions yet</p>
           </div>

apps/web/src/spa/pages/projects.spec.tsx

@@ -51,6 +51,7 @@ afterEach(async () => {
   document.body.replaceChildren();
   root = null;
   apiMock.mockReset();
+  sessionStorage.clear();
 });
 
 async function renderProjectsPage(): Promise<ReturnType<typeof createMemoryRouter>> {
@@ -71,6 +72,22 @@ async function renderProjectsPage(): Promise<ReturnType<typeof createMemoryRoute
   return router;
 }
 
+function clickButtonByText(text: string): void {
+  const button = [...container.querySelectorAll('button')].find((candidate) =>
+    candidate.textContent?.includes(text),
+  );
+  if (!button) {
+    throw new Error(`Button containing "${text}" not found`);
+  }
+  button.dispatchEvent(new MouseEvent('click', { bubbles: true }));
+}
+
+async function flushAct(): Promise<void> {
+  await act(async () => {
+    await Promise.resolve();
+  });
+}
+
 describe('ProjectsPage', () => {
   it('shows a visible loading state while the project request is in flight', async () => {
     const deferred = createDeferred<typeof projectFixtures>();
@@ -91,7 +108,7 @@ describe('ProjectsPage', () => {
 
     const router = await renderProjectsPage();
 
-    expect(apiMock).toHaveBeenCalledWith('/api/projects');
+    expect(apiMock.mock.calls[0]?.[0]).toBe('/api/projects');
     expect(container.textContent).toContain('Mosaic Stack');
     expect(container.textContent).toContain('Agent Runtime');
 
@@ -108,7 +125,7 @@ describe('ProjectsPage', () => {
     expect(container.textContent).toContain('Project detail target');
   });
 
-  it('renders the empty state when the API returns no projects', async () => {
+  it('renders the empty state only for a verified empty collection', async () => {
     apiMock.mockResolvedValueOnce([]);
 
     await renderProjectsPage();
@@ -117,9 +134,12 @@ describe('ProjectsPage', () => {
     expect(container.textContent).toContain(
       'Projects will appear here when created via the gateway API',
     );
+    expect(container.querySelector('[data-freshness]')?.getAttribute('data-freshness')).toBe(
+      'current',
+    );
   });
 
-  it('renders a visible alert when the projects request fails', async () => {
+  it('renders a failed fetch as an explicit unavailable state, never an empty collection', async () => {
     apiMock.mockRejectedValueOnce(new Error('Projects are unavailable'));
 
     await renderProjectsPage();
@@ -127,5 +147,51 @@ describe('ProjectsPage', () => {
     const alert = container.querySelector('[role="alert"]');
     expect(alert).toBeTruthy();
     expect(alert?.textContent).toContain('Projects are unavailable');
+    expect(alert?.textContent).toContain('not an empty result');
+
+    // Negative controls: no healthy empty state and no project cards render
+    // from a failed fetch.
+    expect(container.textContent).not.toContain('No projects yet');
+    expect(container.textContent).not.toContain('Mosaic Stack');
+    expect(container.querySelector('[data-freshness]')?.getAttribute('data-freshness')).toBe(
+      'unavailable',
+    );
+  });
+
+  it('renders an auth failure as unavailable and recovers after retry', async () => {
+    apiMock
+      .mockRejectedValueOnce(Object.assign(new Error('Unauthorized'), { statusCode: 401 }))
+      .mockResolvedValueOnce(projectFixtures);
+
+    await renderProjectsPage();
+
+    const alert = container.querySelector('[role="alert"]');
+    expect(alert?.textContent).toContain('Unauthorized');
+    expect(container.textContent).not.toContain('No projects yet');
+
+    await act(async () => {
+      clickButtonByText('Retry');
+    });
+    await flushAct();
+
+    expect(container.querySelector('[role="alert"]')).toBeNull();
+    expect(container.textContent).toContain('Mosaic Stack');
+    expect(container.querySelector('[data-freshness]')?.getAttribute('data-freshness')).toBe(
+      'current',
+    );
+  });
+
+  it('renders a schema-mismatched response as unavailable, never as data', async () => {
+    apiMock.mockResolvedValueOnce({ results: projectFixtures });
+
+    await renderProjectsPage();
+
+    const alert = container.querySelector('[role="alert"]');
+    expect(alert?.textContent).toContain('not an empty result');
+    expect(container.textContent).not.toContain('Mosaic Stack');
+    expect(container.textContent).not.toContain('No projects yet');
+    expect(container.querySelector('[data-freshness]')?.getAttribute('data-freshness')).toBe(
+      'unavailable',
+    );
   });
 });

apps/web/src/spa/pages/projects.tsx

@@ -1,53 +1,51 @@
-import { useEffect, useState, type ReactElement } from 'react';
+import { type ReactElement } from 'react';
 import { useNavigate } from 'react-router-dom';
 import { ProjectCard } from '@/components/projects/project-card';
+import { StaleDataNotice, UnavailableDataNotice } from '@/components/freshness/freshness-notices';
 import { api } from '@/lib/api';
 import type { Project } from '@/lib/types';
-import { getErrorMessage } from './page-errors';
+import { useFreshCollection, describeFailure } from '@/lib/freshness/use-fresh-collection';
+import { validateProjectCollection } from '@/lib/freshness/validators';
 
 export function ProjectsPage(): ReactElement {
   const navigate = useNavigate();
-  const [projects, setProjects] = useState<Project[]>([]);
+  const projects = useFreshCollection<Project[]>({
-  const [loading, setLoading] = useState(true);
+    source: 'gateway:/api/projects',
-  const [error, setError] = useState<string | null>(null);
+    fetcher: (signal) => api<unknown>('/api/projects', { signal }),
-
+    validate: validateProjectCollection,
-  useEffect(() => {
+    // Projects carry workspace identity (userId) that is only knowable from
-    let cancelled = false;
+    // the payload itself, so a restored entry cannot be scope-checked before
-
+    // display. Conservative choice: no last-known restore for this surface;
-    void api<Project[]>('/api/projects')
+    // cross-workspace switching is still invalidated at verification time.
-      .then((response) => {
+  });
-        if (cancelled) return;
+  const retry = (): void => {
-        setProjects(response);
+    void projects.revalidate();
-      })
+  };
-      .catch((caught: unknown) => {
-        if (cancelled) return;
-        setError(getErrorMessage(caught, 'Failed to load projects.'));
-      })
-      .finally(() => {
-        if (cancelled) return;
-        setLoading(false);
-      });
-
-    return () => {
-      cancelled = true;
-    };
-  }, []);
 
   return (
-    <div className="flex min-h-screen flex-col px-4 py-6 sm:px-6">
+    <div
+      data-freshness={projects.freshness}
+      className="flex min-h-screen flex-col px-4 py-6 sm:px-6"
+    >
       <header className="mb-6 border-b px-1 pb-3">
         <h1 className="text-2xl font-semibold">Projects</h1>
       </header>
 
-      {error ? (
+      {projects.freshness === 'stale' && projects.snapshot ? (
-        <div role="alert" className="mb-6 rounded-lg border border-error/40 px-4 py-3 text-sm">
+        <div className="mb-6">
-          {error}
+          <StaleDataNotice label={projects.snapshot} onRetry={retry} />
         </div>
       ) : null}
 
-      {loading ? (
+      {projects.freshness === 'unknown' ? (
         <p className="py-8 text-center text-sm text-text-muted">Loading projects...</p>
-      ) : projects.length === 0 ? (
+      ) : projects.freshness === 'unavailable' ? (
+        <UnavailableDataNotice
+          title="Projects"
+          detail={describeFailure(projects.failure)}
+          onRetry={retry}
+        />
+      ) : projects.data !== null && projects.data.length === 0 ? (
         <div className="py-12 text-center">
           <h2 className="text-lg font-medium text-text-secondary">No projects yet</h2>
           <p className="mt-1 text-sm text-text-muted">
@@ -56,7 +54,7 @@ export function ProjectsPage(): ReactElement {
         </div>
       ) : (
         <div className="grid gap-4 sm:grid-cols-2 lg:grid-cols-3">
-          {projects.map((project) => (
+          {(projects.data ?? []).map((project) => (
             <ProjectCard
               key={project.id}
               project={project}

apps/web/src/spa/pages/tasks.spec.tsx

@@ -3,6 +3,9 @@ import { createRoot, type Root } from 'react-dom/client';
 import { createMemoryRouter, RouterProvider, type RouteObject } from 'react-router-dom';
 import { afterAll, afterEach, beforeAll, describe, expect, it, vi } from 'vitest';
 import { taskFixtures } from './page-fixtures';
+import { acceptSnapshot, DEFAULT_FRESHNESS_POLICY } from '@/lib/freshness/model';
+import { writeSnapshotCache } from '@/lib/freshness/snapshot-cache';
+import { validateTaskCollection } from '@/lib/freshness/validators';
 
 const { apiMock } = vi.hoisted(() => ({
   apiMock: vi.fn(),
@@ -48,6 +51,7 @@ afterEach(async () => {
   document.body.replaceChildren();
   root = null;
   apiMock.mockReset();
+  sessionStorage.clear();
 });
 
 async function renderTasksPage(): Promise<void> {
@@ -72,6 +76,13 @@ function clickButtonByText(text: string): void {
   button.dispatchEvent(new MouseEvent('click', { bubbles: true }));
 }
 
+/** Flush pending promise callbacks inside the act environment. */
+async function flushAct(): Promise<void> {
+  await act(async () => {
+    await Promise.resolve();
+  });
+}
+
 describe('TasksPage', () => {
   it('shows a visible loading state before the tasks request settles', async () => {
     const deferred = createDeferred<typeof taskFixtures>();
@@ -132,7 +143,7 @@ describe('TasksPage', () => {
     expect(container.textContent).toContain('Wire list and kanban modal interactions');
   });
 
-  it('renders a visible alert when the tasks request fails', async () => {
+  it('renders a failed fetch as an explicit unavailable state, never an empty healthy board', async () => {
     apiMock.mockRejectedValueOnce(new Error('Tasks request failed'));
 
     await renderTasksPage();
@@ -140,5 +151,80 @@ describe('TasksPage', () => {
     const alert = container.querySelector('[role="alert"]');
     expect(alert).toBeTruthy();
     expect(alert?.textContent).toContain('Tasks request failed');
+    expect(alert?.textContent).toContain('not an empty result');
+
+    // Negative controls: no board, no healthy empty-state markers, and the
+    // surface is marked unavailable rather than current.
+    expect(container.textContent).not.toContain('Not Started');
+    expect(container.textContent).not.toContain('No tasks');
+    expect(container.querySelector('[data-freshness]')?.getAttribute('data-freshness')).toBe(
+      'unavailable',
+    );
+  });
+
+  it('recovers to a current board after retrying a failed fetch', async () => {
+    apiMock
+      .mockRejectedValueOnce(new Error('Tasks request failed'))
+      .mockResolvedValueOnce(taskFixtures);
+
+    await renderTasksPage();
+    expect(container.querySelector('[role="alert"]')).toBeTruthy();
+
+    await act(async () => {
+      clickButtonByText('Retry');
+    });
+    await flushAct();
+
+    expect(container.querySelector('[role="alert"]')).toBeNull();
+    expect(container.textContent).toContain('Not Started');
+    expect(container.querySelector('[data-freshness]')?.getAttribute('data-freshness')).toBe(
+      'current',
+    );
+  });
+
+  it('labels restored last-known data as stale with source, version, and age until verified', async () => {
+    // Seed a last-known snapshot fetched five minutes ago; the page must
+    // render it only under an explicit staleness label while the fetch is
+    // still in flight.
+    const restored = acceptSnapshot({
+      value: taskFixtures,
+      validate: validateTaskCollection,
+      previous: null,
+      policy: DEFAULT_FRESHNESS_POLICY,
+      source: 'gateway:/api/tasks',
+      now: Date.now() - 5 * 60_000,
+    });
+    if (restored.outcome !== 'accepted') throw new Error('fixture setup failed');
+    writeSnapshotCache('tasks', restored.snapshot);
+
+    const deferred = createDeferred<typeof taskFixtures>();
+    apiMock.mockReturnValueOnce(deferred.promise);
+
+    await renderTasksPage();
+
+    expect(container.querySelector('[data-freshness]')?.getAttribute('data-freshness')).toBe(
+      'stale',
+    );
+    const banner = container.querySelector('[role="status"]');
+    expect(banner?.textContent).toContain('last-known');
+    expect(banner?.textContent).toContain('may be out of date');
+    expect(banner?.textContent).toContain('gateway:/api/tasks');
+    expect(banner?.textContent).toContain('snapshot v1');
+    expect(banner?.textContent).toContain('5m ago');
+
+    // Last-known data still renders as situational awareness under the label.
+    expect(container.textContent).toContain('Route /tasks');
+    expect(container.textContent).not.toContain('Loading tasks...');
+
+    // Verification lands: the banner clears and the surface becomes current.
+    await act(async () => {
+      deferred.resolve(taskFixtures);
+      await deferred.promise;
+    });
+
+    expect(container.querySelector('[role="status"]')).toBeNull();
+    expect(container.querySelector('[data-freshness]')?.getAttribute('data-freshness')).toBe(
+      'current',
+    );
   });
 });

apps/web/src/spa/pages/tasks.tsx

@@ -1,45 +1,32 @@
-import { useEffect, useState, type ReactElement } from 'react';
+import { useState, type ReactElement } from 'react';
 import { KanbanBoard } from '@/components/tasks/kanban-board';
 import { TaskDetailModal } from '@/components/tasks/task-detail-modal';
 import { TaskListView } from '@/components/tasks/task-list-view';
+import { StaleDataNotice, UnavailableDataNotice } from '@/components/freshness/freshness-notices';
 import { api } from '@/lib/api';
 import { cn } from '@/lib/cn';
 import type { Task } from '@/lib/types';
-import { getErrorMessage } from './page-errors';
+import { useFreshCollection, describeFailure } from '@/lib/freshness/use-fresh-collection';
+import { validateTaskCollection } from '@/lib/freshness/validators';
 
 type ViewMode = 'list' | 'kanban';
 
 export function TasksPage(): ReactElement {
-  const [tasks, setTasks] = useState<Task[]>([]);
+  const tasks = useFreshCollection<Task[]>({
+    source: 'gateway:/api/tasks',
+    fetcher: (signal) => api<unknown>('/api/tasks', { signal }),
+    validate: validateTaskCollection,
+    cacheKey: 'tasks',
+  });
   const [view, setView] = useState<ViewMode>('kanban');
-  const [loading, setLoading] = useState(true);
-  const [error, setError] = useState<string | null>(null);
   const [selectedTask, setSelectedTask] = useState<Task | null>(null);
 
-  useEffect(() => {
+  const retry = (): void => {
-    let cancelled = false;
+    void tasks.revalidate();
-
+  };
-    void api<Task[]>('/api/tasks')
-      .then((response) => {
-        if (cancelled) return;
-        setTasks(response);
-      })
-      .catch((caught: unknown) => {
-        if (cancelled) return;
-        setError(getErrorMessage(caught, 'Failed to load tasks.'));
-      })
-      .finally(() => {
-        if (cancelled) return;
-        setLoading(false);
-      });
-
-    return () => {
-      cancelled = true;
-    };
-  }, []);
 
   return (
-    <div className="flex min-h-screen flex-col px-4 py-6 sm:px-6">
+    <div data-freshness={tasks.freshness} className="flex min-h-screen flex-col px-4 py-6 sm:px-6">
       <header className="mb-6 flex items-center justify-between gap-4 border-b px-1 pb-3">
         <h1 className="text-2xl font-semibold">Tasks</h1>
         <div className="flex rounded-lg border border-surface-border">
@@ -70,18 +57,24 @@ export function TasksPage(): ReactElement {
         </div>
       </header>
 
-      {error ? (
+      {tasks.freshness === 'stale' && tasks.snapshot ? (
-        <div role="alert" className="mb-6 rounded-lg border border-error/40 px-4 py-3 text-sm">
+        <div className="mb-6">
-          {error}
+          <StaleDataNotice label={tasks.snapshot} onRetry={retry} />
         </div>
       ) : null}
 
-      {loading ? (
+      {tasks.freshness === 'unknown' ? (
         <p className="py-8 text-center text-sm text-text-muted">Loading tasks...</p>
+      ) : tasks.freshness === 'unavailable' ? (
+        <UnavailableDataNotice
+          title="Tasks"
+          detail={describeFailure(tasks.failure)}
+          onRetry={retry}
+        />
       ) : view === 'kanban' ? (
-        <KanbanBoard tasks={tasks} onTaskClick={setSelectedTask} />
+        <KanbanBoard tasks={tasks.data ?? []} onTaskClick={setSelectedTask} />
       ) : (
-        <TaskListView tasks={tasks} onTaskClick={setSelectedTask} />
+        <TaskListView tasks={tasks.data ?? []} onTaskClick={setSelectedTask} />
       )}
 
       {selectedTask ? (

docs/release-integrity/TASKS.md

@@ -15,18 +15,18 @@
 > `done` requires: repo quality gates green, independent review recorded, terminal-green CI on
 > the PR head, squash merge to `next`, and acceptance evidence in notes.
 
-| id       | status      | description                                                                                                                                                                                                                                                                      | issue | agent      | repo              | branch                            | depends_on                                                 | estimate | notes                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                              |
+| id       | status      | description                                                                                                                                                                                                                                                                      | issue | agent      | repo              | branch                            | depends_on                                                 | estimate | notes                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                           |
-| -------- | ----------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----- | ---------- | ----------------- | --------------------------------- | ---------------------------------------------------------- | -------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| -------- | ----------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----- | ---------- | ----------------- | --------------------------------- | ---------------------------------------------------------- | -------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| RI-0-001 | in-progress | Bootstrap: issue #1275, PRD section, this DAG, scratchpad (docs only)                                                                                                                                                                                                            | #1275 | pi-glm-5.3 | mosaicstack/stack | docs/ri-050-mission-bootstrap     | —                                                          | 6K       |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
+| RI-0-001 | done        | Bootstrap: issue #1275, PRD section, this DAG, scratchpad (docs only)                                                                                                                                                                                                            | #1275 | pi-glm-5.3 | mosaicstack/stack | docs/ri-050-mission-bootstrap     | —                                                          | 6K       | PR #1276 (head 758659dd): docs-only, CI green (2475). Review requested from fargo. Merges first (no publish run).                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                               |
-| RI-1-001 | in-progress | RI-N1: canonical terminal verification command + publish-pipeline exact-commit gate (every publish step depends on verify; commit identity check; fail closed)                                                                                                                   | #1275 | pi-glm-5.3 | mosaicstack/stack | feat/ri-050-publish-gate          | RI-0-001                                                   | 25K      |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
+| RI-1-001 | done        | RI-N1: canonical terminal verification command + publish-pipeline exact-commit gate (every publish step depends on verify; commit identity check; fail closed)                                                                                                                   | #1275 | pi-glm-5.3 | mosaicstack/stack | feat/ri-050-publish-gate          | RI-0-001                                                   | 25K      | PR #1277 (head 46784c8d): CI GREEN at head after serialized retry (pipeline 2476, 2026-08-18) - earlier red was CI-agent contention (web SPA timeouts under concurrent pipelines), not code. Review requested from fargo at pinned head (comms 20260818T021025Z).                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                               |
-| RI-1-002 | not-started | RI-N1 negative control: checked-in tests proving a broken mandatory check blocks every publish step and that DAG edges cannot be bypassed                                                                                                                                        | #1275 | pi-glm-5.3 | mosaicstack/stack | test/ri-050-publish-gate-negative | RI-1-001                                                   | 12K      |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
+| RI-1-002 | done        | RI-N1 negative control: checked-in tests proving a broken mandatory check blocks every publish step and that DAG edges cannot be bypassed                                                                                                                                        | #1275 | pi-glm-5.3 | mosaicstack/stack | test/ri-050-publish-gate-negative | RI-1-001                                                   | 12K      |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 |
-| RI-2-001 | in-progress | RI-N2 (Forge): remove stub-executor false success; `--simulate` typed `simulated` results that satisfy nothing; literal-`true` gates and echo-review replaced with real gates or typed waiting-for-authority                                                                     | #1275 | pi-glm-5.3 | mosaicstack/stack | fix/ri-050-forge-fail-closed      | RI-0-001                                                   | 20K      | Independent review APPROVED 2026-08-17 (Gitea review 172 on PR #1278, head 99b8f6ea; reviewing seat fargo — recorded under shared host principal mos-dt-0, provenance correction posted by fred; wrapper gap filed by fred). Executed at head: forge tests 116/116, lint green, typecheck green after building macp dist (minimal-install artifact, not a defect), workspace typecheck 45/45, no external type consumers of the changed interfaces. CI red = known lane-wide fleet-test failure only, carries no information about this change (fred, log-content analysis, pipelines 2456-2458). Non-blocking finding: README L141-143 + skills/mosaic-forge/SKILL.md document bare forge run/resume, which now fails closed — fast-follow docs touch. Merge queued behind #1270. |
+| RI-2-001 | done        | RI-N2 (Forge): remove stub-executor false success; `--simulate` typed `simulated` results that satisfy nothing; literal-`true` gates and echo-review replaced with real gates or typed waiting-for-authority                                                                     | #1275 | pi-glm-5.3 | mosaicstack/stack | fix/ri-050-forge-fail-closed      | RI-0-001                                                   | 20K      | Independent review APPROVED 2026-08-17 (Gitea review 172 on PR #1278, head 99b8f6ea; reviewing seat fargo — recorded under shared host principal mos-dt-0, provenance correction posted by fred; wrapper gap filed by fred). Executed at head: forge tests 116/116, lint green, typecheck green after building macp dist (minimal-install artifact, not a defect), workspace typecheck 45/45, no external type consumers of the changed interfaces. CI red = known lane-wide fleet-test failure only, carries no information about this change (fred, log-content analysis, pipelines 2456-2458). Non-blocking finding: README L141-143 + skills/mosaic-forge/SKILL.md document bare forge run/resume, which now fails closed — fast-follow docs touch. Merge queued behind #1270. UPDATE 2026-08-18: #1270 merged; CI GREEN at head 4917df1f via serialized retry (pipeline 2477) - root cause of prior reds was CI-agent contention (web SPA timeouts under concurrent pipelines), superseding the fleet-test-failure theory. |
-| RI-2-002 | in-progress | RI-N2 (MACP): gate runner fails closed on empty commands, stub executors, and unimplemented CI-provider gates unless explicit simulate; typed capability failures                                                                                                                | #1275 | pi-glm-5.3 | mosaicstack/stack | fix/ri-050-macp-fail-closed       | RI-0-001                                                   | 15K      |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
+| RI-2-002 | done        | RI-N2 (MACP): gate runner fails closed on empty commands, stub executors, and unimplemented CI-provider gates unless explicit simulate; typed capability failures                                                                                                                | #1275 | pi-glm-5.3 | mosaicstack/stack | fix/ri-050-macp-fail-closed       | RI-0-001                                                   | 15K      | PR #1293 (head 2097379e): CI green (pipeline 2465), independent review APPROVED (Gitea review 173, jarvis seat, 2026-08-17) - macp 109/109 verified at head. Merge queued behind #1276/#1277/#1278.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                             |
-| RI-3-001 | not-started | RI-N4: complete probe inventory mapping every TS and shell quality-rail check to one canonical check with disposition (preserve/strengthen/retire, each named)                                                                                                                   | #1275 | pi-glm-5.3 | mosaicstack/stack | docs/ri-050-qr-probe-inventory    | RI-0-001                                                   | 12K      |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
+| RI-3-001 | done        | RI-N4: complete probe inventory mapping every TS and shell quality-rail check to one canonical check with disposition (preserve/strengthen/retire, each named)                                                                                                                   | #1275 | pi-glm-5.3 | mosaicstack/stack | docs/ri-050-qr-probe-inventory    | RI-0-001                                                   | 12K      | PR #1302 (head e06a47fac591): CI green (2484), independent review APPROVED (Gitea review 187, fargo seat, 2026-08-18) — 54 rows / 21 canonical checks / dispositions 43-2-9-0 verified by row-count and code spot-checks. Merged by fargo at pinned head.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       |
-| RI-3-002 | not-started | RI-N4: TS evaluator absorbs effective shell probes; typed results (passed/failed/blocked/error/not-applicable) with versioned digested check definitions; shell commands become thin adapters; contract/parity/negative-control tests                                            | #1275 | pi-glm-5.3 | mosaicstack/stack | feat/ri-050-qr-evaluator          | RI-3-001                                                   | 30K      |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
+| RI-3-002 | not-started | RI-N4: TS evaluator absorbs effective shell probes; typed results (passed/failed/blocked/error/not-applicable) with versioned digested check definitions; shell commands become thin adapters; contract/parity/negative-control tests                                            | #1275 | pi-glm-5.3 | mosaicstack/stack | feat/ri-050-qr-evaluator          | RI-3-001                                                   | 30K      |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 |
-| RI-4-001 | in-progress | RI-N3: one PRD application service — `mission --plan` persists mission↔PRD linkage (ids/versions/selected requirements); `mosaic prdy` routes through the service or becomes a named import/export adapter; Markdown is a labeled generated view; explicit conflict-aware import | #1275 | pi-glm-5.3 | mosaicstack/stack | feat/ri-050-prd-authority         | RI-0-001                                                   | 35K      |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
+| RI-4-001 | in-progress | RI-N3: one PRD application service — `mission --plan` persists mission↔PRD linkage (ids/versions/selected requirements); `mosaic prdy` routes through the service or becomes a named import/export adapter; Markdown is a labeled generated view; explicit conflict-aware import | #1275 | pi-glm-5.3 | mosaicstack/stack | feat/ri-050-prd-authority         | RI-0-001                                                   | 35K      | PR #1294 (head 8d258e1d): CI green (pipeline 2466), independent review APPROVED (Gitea review 174, jarvis seat, 2026-08-17) - prdy 20/20 + command specs 9/9 at head. Merge queued behind #1276/#1277/#1278.                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
-| RI-5-001 | not-started | RI-N5: typed freshness states (current/stale/partial/unknown/unavailable); no failed-fetch-renders-empty; stale derived verdicts → unknown; mutations disabled when stale; failure-matrix tests                                                                                  | #1275 | pi-glm-5.3 | mosaicstack/stack | feat/ri-050-web-stale-safety      | RI-0-001                                                   | 25K      |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
+| RI-5-001 | done        | RI-N5: typed freshness states (current/stale/partial/unknown/unavailable); no failed-fetch-renders-empty; stale derived verdicts → unknown; mutations disabled when stale; failure-matrix tests                                                                                  | #1275 | pi-glm-5.3 | mosaicstack/stack | feat/ri-050-web-stale-safety      | RI-0-001                                                   | 25K      |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 |
-| RI-V-001 | not-started | Final verification + release evidence: all cards verified merged, negative controls demonstrated, real `next` publish run green on exact commit, evidence pack recorded                                                                                                          | #1275 | pi-glm-5.3 | mosaicstack/stack | docs/ri-050-release-evidence      | RI-1-002, RI-2-001, RI-2-002, RI-3-002, RI-4-001, RI-5-001 | 10K      |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
+| RI-V-001 | not-started | Final verification + release evidence: all cards verified merged, negative controls demonstrated, real `next` publish run green on exact commit, evidence pack recorded                                                                                                          | #1275 | pi-glm-5.3 | mosaicstack/stack | docs/ri-050-release-evidence      | RI-1-002, RI-2-001, RI-2-002, RI-3-002, RI-4-001, RI-5-001 | 10K      |                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 |
 
 ## Dispatch waves (max 2 parallel workers)
 

docs/release-integrity/probe-inventory.md

@@ -0,0 +1,186 @@
+# Quality-Rails Probe Inventory — RI-3-001
+
+- **Task:** RI-3-001 (SDLC-D-037 first half; PRD § Release Integrity Workstream, RI-N4)
+- **Date:** 2026-08-18
+- **Base:** `origin/next` @ `8199261c` (branch `docs/ri-050-qr-probe-inventory`)
+- **Follow-up:** RI-3-002 consumes the dispositions here when building the single TS evaluator.
+
+## 0. Scope and method
+
+Every mechanism in this repository that verifies a quality, integrity, safety, or release
+property — TypeScript checks, shell probes, pipeline steps, git hooks, and installer-side
+assertions — gets one row. Each row's "what it actually verifies" was written from the
+probe's **code**, not its name or docs. Framework tool unit/regression suites (git wrappers,
+wake, tmux, orchestrator, …) are treated as one enforcement surface (`test:framework-shell`)
+because they test tool behavior rather than repo quality; their wiring integrity is itself
+guarded by `check-test-enumeration.sh`, and the quality-relevant members are rowed
+individually.
+
+**Kinds:** `ts` (TypeScript/Node check), `shell` (bash/python probe), `pipeline-step`
+(exists only inside a Woodpecker pipeline).
+
+**Enforcement points:** `local` (operator-invoked), `pre-commit`, `pre-push`,
+`CI ci.yml#<step>`, `publish.yml#<step>` (CI on push to main/next), `turbo <task>`,
+`agent-runtime` (framework hooks on an agent host), `installer` (host install path),
+`unwired`.
+
+**Dispositions** (recommendations for RI-3-002): `preserve` (keep as-is; already the
+canonical or a correct guard-of-the-guard), `strengthen` (keep, but a concrete gap must
+close — usually absorption into the TS evaluator), `strengthen (review)` (viable retirement
+candidate once the evaluator absorbs it; do not retire yet). Note: RI-N4 requires that
+effective shell probes be **absorbed before** their independent paths retire — no row here
+is marked `retire` because no absorption exists yet.
+
+## 1. Inventory
+
+### 1.1 Repo-level gate tasks (pnpm / turbo)
+
+| check                                 | location                                                                             | kind | what it actually verifies                                                                                                                                                                                                                                                                                                                                                                                                     | enforcement point                                                                                      | canonical check           | disposition | rationale                                                                                                                                                 |
+| ------------------------------------- | ------------------------------------------------------------------------------------ | ---- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------ | ------------------------- | ----------- | --------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `pnpm preflight` (checkout preflight) | `scripts/preflight.mjs`                                                              | ts   | Six gate binaries (eslint, husky, prettier, tsc, turbo, vitest) exist and are executable in `node_modules/.bin` (exit 42 if not); no stale `.mosaic-test-work/web-build.lock` (exit 43); `apps/web/.next` is a real directory (not a symlink), every entry owned by the current uid, and its `.mosaic-source-hash` fingerprint + `.mosaic-symlink-manifest` hash match the certified build written by `scripts/build-web.mjs` | `pre-push`; inside `pnpm typecheck` (→ `CI ci.yml#typecheck`, verify-release `typecheck` stage)        | QC-1 Checkout integrity   | preserve    | Blocks a poisoned/stale generated `.next` from faking a green typecheck (the five-month-stale-`.next` class); trust chain is self-contained per-checkout. |
+| `pnpm typecheck`                      | root `package.json` → `turbo run typecheck`                                          | ts   | Per-package `tsc --noEmit` (all 20 packages); turbo `typecheck` depends on `^build`, so package builds must succeed first; prefixed by checkout preflight                                                                                                                                                                                                                                                                     | `CI ci.yml#typecheck`; `pre-push`; verify-release `typecheck` stage; `turbo typecheck`                 | QC-2 Workspace typecheck  | preserve    | The single workspace-wide type gate; CI and hooks invoke the same task, no divergent checklist.                                                           |
+| `pnpm lint`                           | root `package.json` → `turbo run lint`                                               | ts   | Per-package `eslint src` under root `eslint.config.mjs` (ignores `dist`, `.next`, `framework/**`, etc.)                                                                                                                                                                                                                                                                                                                       | `CI ci.yml#lint`; `pre-push`; verify-release `lint` stage; `turbo lint`                                | QC-3 Workspace lint       | preserve    | Same-task invocation from every surface; no second lint definition.                                                                                       |
+| `pnpm format:check`                   | root `package.json` → `prettier --check`                                             | ts   | Prettier parse/format equality over `**/*.{ts,tsx,js,jsx,json,md}` minus `.prettierignore` (generated trees, `docs/scratchpads/`, venvs, …)                                                                                                                                                                                                                                                                                   | `CI ci.yml#format`; `pre-push`; verify-release `format` stage                                          | QC-4 Format check         | preserve    | Single formatter, single ignore list, enforced identically everywhere.                                                                                    |
+| `pnpm test`                           | root `package.json` `test` = `test:checkout` && `turbo run test` && `test:installer` | ts   | (a) `node --test scripts/*.test.mjs` — checkout-tool units; (b) per-package `vitest run` (mosaic appends the 47-command `test:framework-shell` chain); (c) `tools/install-next-lane.test.sh`; turbo `test` declares DB env vars and depends on `^build`                                                                                                                                                                       | `CI ci.yml#test` (with `DATABASE_URL` + `db:migrate` first); verify-release `test` stage; `turbo test` | QC-5 Test suite execution | preserve    | One composed test command; the chain property (any link red ⇒ step red) is the gate.                                                                      |
+| `pnpm build`                          | root `package.json` → `turbo run build`                                              | ts   | Per-package build (`tsc`/Next) with `^build` dependency and `dist/**` outputs                                                                                                                                                                                                                                                                                                                                                 | `publish.yml#build`; verify-release `build` stage; `turbo build`                                       | QC-6 Workspace build      | preserve    | Publish artifacts derive from the same build task CI verifies.                                                                                            |
+
+### 1.2 Framework quality shell probes (`packages/mosaic/framework/tools/quality/`)
+
+| check                                       | location                                                                | kind  | what it actually verifies                                                                                                                                                                                                                                                                                                                                                                                                                                                                        | enforcement point                                                                                                                                                                                                                                                                                                                                                           | canonical check                                                                                                                         | disposition                      | rationale                                                                                                                                                                                                                                                         |
+| ------------------------------------------- | ----------------------------------------------------------------------- | ----- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------- |
+| Sanitization gate                           | `scripts/verify-sanitized.sh`                                           | shell | Built-in self-test first (planted identity/structural/YAML+service fixtures; exit 2 if the regexes or extension coverage break), then: (1) identity denylist grep (`jarvis\|jason\|woltje\|brain.woltje.com\|/home/jwoltje\|\bPDA\b`) over all shipped text files **including** `examples/`; (2) structural grep for private `$HOME/src` defaults in shipped scripts **excluding** `examples/`. Any hit ⇒ exit 1                                                                                 | `CI ci.yml#sanitization`; verify-release `sanitization` stage                                                                                                                                                                                                                                                                                                               | QC-7 Framework sanitization                                                                                                             | preserve                         | Labeled one-time regression guard with a self-test that prevents silent no-op; correctly scoped (identity vs structural) and documented as not a general PII detector.                                                                                            |
+| Resident-context budget                     | `scripts/check-resident-budget.sh` (+ `--self-test`)                    | shell | Self-test of the comparator, then `wc -l` vs per-file ceilings (CONSTITUTION 120, AGENTS 120, each RUNTIME.md 90); missing file ⇒ fail; over ceiling ⇒ exit 1                                                                                                                                                                                                                                                                                                                                    | `CI ci.yml#sanitization` (both modes); verify-release `sanitization` stage                                                                                                                                                                                                                                                                                                  | QC-8 Resident-context budget                                                                                                            | preserve                         | Caps the container (lines), never the wording — the deliberate anti-drift design (DESIGN §7); CI-enforceable half only, by design.                                                                                                                                |
+| Test-membership enumeration guard (#1017)   | `scripts/check-test-enumeration.sh` + `test-enumeration-exclusions.txt` | shell | Parses surface S1 (`packages/mosaic` `test:framework-shell` via JSON+shlex) and S2 (every `framework/tools/\*.sh                                                                                                                                                                                                                                                                                                                                                                                 | .py`token in`ci.yml`, comment lines stripped); population = `_test_.sh`under`framework/tools`; FAILS on: suite-shaped file on disk neither enumerated nor signed-excluded; surface naming a path missing on disk (both directions); exclusion without reason / stale / outside population / contradicting enumeration. Proves **naming, not reachability** (stated in-file) | `CI ci.yml#sanitization` (direct line); link [0] of `test:framework-shell` (thus `CI ci.yml#test`); verify-release `sanitization` stage | QC-9 Test-membership enumeration | preserve                                                                                                                                                                                                                                                          | Makes silent under-run impossible; invoked from both surfaces it audits so severing the chain cannot silence it. |
+| Enumeration-guard needles                   | `scripts/test-check-test-enumeration.sh`                                | shell | Needle/control fixtures driven through `--root`: every promised failure mode must trip the guard **on its own words**, plus controls that must pass (null-case defense); covers commented-out ci.yml lines (F1) and line-range parsing (n2b)                                                                                                                                                                                                                                                     | `test:framework-shell` → `CI ci.yml#test`; verify-release `test` stage                                                                                                                                                                                                                                                                                                      | QC-9 Test-membership enumeration                                                                                                        | preserve                         | Guard-of-the-guard with both polarities; same canonical check by design.                                                                                                                                                                                          |
+| Upgrade manifest guard (#791 HARD GATE)     | `scripts/test-upgrade-manifest-guard.sh`                                | shell | Keep-mode `install.sh` upgrade against seeded throwaway `MOSAIC_HOME`: every operator sentinel — including an **unanticipated** one — survives byte-identical with unchanged mtime; framework files still update; retired framework files pruned; matrix run with rsync present AND absent (keep path must be rsync-independent); fail-closed matrix (empty/operator-only/malformed/missing manifest aborts loudly, operator files untouched); operator secret never appears in installer output | `CI ci.yml#upgrade-guard`; verify-release `upgrade-guard` stage                                                                                                                                                                                                                                                                                                             | QC-10 Upgrade/install safety                                                                                                            | preserve                         | The operator-data hard gate for the `mosaic update` path; negative controls are load-bearing and documented.                                                                                                                                                      |
+| Upgrade rollback gate (#791 B1)             | `scripts/test-upgrade-rollback.sh`                                      | shell | Mid-sync failure (PATH-shadowing `cp` shim) must trigger snapshot restore: restore message fires, corrupted file restored, target byte-identical to pre-upgrade; control installer with `set -E` stripped must NOT roll back (proves errtrace is load-bearing); plus signal/exit-guard controls                                                                                                                                                                                                  | `CI ci.yml#upgrade-guard`; verify-release `upgrade-guard` stage                                                                                                                                                                                                                                                                                                             | QC-10 Upgrade/install safety                                                                                                            | preserve                         | Proves the rollback trap actually fires; the `-E`-stripped control keeps Part A honest.                                                                                                                                                                           |
+| Durable-snapshot gate (#791 PR2)            | `scripts/test-upgrade-durable-snapshot.sh`                              | shell | Pre-update snapshot taken before any mutation (0700/0600 perms, secret never logged, retention-pruned); post-sync verify net restores operator files a manifest bug lets the sync touch; CWE-59 symlink-leaf guard proven with a portable cp shim in both polarities (write-through-link must not happen); v1→v2 migration semantics (intended `bin/` removal not healed)                                                                                                                        | `CI ci.yml#upgrade-guard`; verify-release `upgrade-guard` stage                                                                                                                                                                                                                                                                                                             | QC-10 Upgrade/install safety                                                                                                            | preserve                         | Covers tampering and leak vectors the manifest guard cannot see; the shim rationale (busybox vs GNU cp) is documented in-file.                                                                                                                                    |
+| Install migration matrix (v2→v3)            | `scripts/test-install-migration.sh`                                     | shell | Fixture matrix running the real installer with `MOSAIC_SYNC_ONLY=1`: fresh install seeds + stamps version 3; legacy user-edited AGENTS overwritten with `.pre-constitution.bak` preserved (and idempotent); tuned STANDARDS overwritten; operator files (SOUL, credentials) preserved. Mirrors the TS suite `packages/mosaic/src/config/file-adapter.test.ts` — both installers must behave identically                                                                                          | `CI ci.yml#upgrade-guard`; verify-release `upgrade-guard` stage                                                                                                                                                                                                                                                                                                             | QC-10 Upgrade/install safety                                                                                                            | preserve                         | Pins the shell/TS installer parity contract; removal would orphan that parity requirement.                                                                                                                                                                        |
+| Enforcement verification probe (bash)       | `scripts/verify.sh`                                                     | shell | Attempts **real commits** in the target repo: planted type error must produce a commit blocked with `error`; planted `any` must trip `no-explicit-any`; planted lint error must trip `prettier`; gitleaks binary must exist (3a) and detect a planted AWS key via `gitleaks git --pre-commit --staged --redact` (3b). Verdicts are output-grep matches on hook stderr                                                                                                                            | `local` via installed `mosaic-quality-verify` on scaffolded target projects; **not run in this repo's CI**                                                                                                                                                                                                                                                                  | QC-20 Downstream enforcement verification                                                                                               | strengthen (review)              | Mechanism is genuinely behavioral (stronger than file presence) but verdict logic is grep-on-output and it is unwired here; absorb as the evaluator's enforcement-probe check (the RI-N4 evaluator invokes it or reimplements it) before retiring the shell path. |
+| Enforcement verification probe (PowerShell) | `scripts/verify.ps1`                                                    | shell | Windows port of `verify.sh`: same planted-commit tests with `$output -match` matching; no gitleaks self-test parity beyond the same checks                                                                                                                                                                                                                                                                                                                                                       | `local` (Windows operator); no Windows CI runner exists                                                                                                                                                                                                                                                                                                                     | QC-20 Downstream enforcement verification                                                                                               | strengthen (review)              | A hand-maintained twin of `verify.sh` with no CI coverage — exactly the drift shape the single evaluator removes; retire after the TS evaluator owns the probe.                                                                                                   |
+| Quality template installer (bash)           | `scripts/install.sh`                                                    | shell | Copies template files (`.husky/pre-commit` incl. mandatory gitleaks, `.lintstagedrc.js`, `.eslintrc.js`, `tsconfig.json`, `.woodpecker.yml`, `.gitleaks.toml`) into a target project; **warns** (does not verify) about `package.json` snippet merge; no post-condition check                                                                                                                                                                                                                    | `local` / via `mosaic-quality-apply`                                                                                                                                                                                                                                                                                                                                        | QC-21 Downstream rails scaffolding                                                                                                      | strengthen (review)              | Duplicates the TS `quality-rails init` scaffolder for a different template set; converging on one scaffolder (with post-scaffold verification) is prerequisite to retiring this path.                                                                             |
+| Quality template installer (PowerShell)     | `scripts/install.ps1`                                                   | shell | Windows twin of the template copy above                                                                                                                                                                                                                                                                                                                                                                                                                                                          | `local` (Windows operator)                                                                                                                                                                                                                                                                                                                                                  | QC-21 Downstream rails scaffolding                                                                                                      | strengthen (review)              | Same twin-drift risk as `verify.ps1`; no runner exercises it.                                                                                                                                                                                                     |
+| `mosaic-quality-verify` adapter             | `framework/tools/_scripts/mosaic-quality-verify`                        | shell | Thin adapter: validates target dir exists, asserts `verify.sh` present+executable, `cd` target, exec it. No verdict logic of its own                                                                                                                                                                                                                                                                                                                                                             | `local` (installed framework bin)                                                                                                                                                                                                                                                                                                                                           | QC-20 Downstream enforcement verification                                                                                               | preserve                         | Already the thin-adapter shape RI-N4 prescribes for shell surfaces.                                                                                                                                                                                               |
+| `mosaic-quality-apply` adapter              | `framework/tools/_scripts/mosaic-quality-apply`                         | shell | Thin adapter: arg validation then exec of quality `install.sh --template … --target …`                                                                                                                                                                                                                                                                                                                                                                                                           | `local` (installed framework bin)                                                                                                                                                                                                                                                                                                                                           | QC-21 Downstream rails scaffolding                                                                                                      | preserve                         | Thin adapter, no separate verdict; disposition follows its target script's convergence.                                                                                                                                                                           |
+| Roster schema regression                    | `scripts/test-roster-schema.py`                                         | shell | jsonschema `Draft202012Validator` over `fleet/roster.schema.json` with valid/invalid connector-kind fixtures (tmux/discord/matrix conditional fields)                                                                                                                                                                                                                                                                                                                                            | **unwired** — not on S1 or S2, not signed-excluded; also outside the enumeration guard's `*.sh` population, so the guard cannot see it                                                                                                                                                                                                                                      | QC-5 Test suite execution                                                                                                               | strengthen (review)              | A real regression suite that currently runs nowhere; wire it into a CI surface or sign an exclusion — leaving it invisible re-arms the exact gap #1017 closed.                                                                                                    |
+| Framework shell chain (S1)                  | `packages/mosaic/package.json` `test:framework-shell`                   | shell | 47-command `&&` chain: enumeration guard + needles, 14 lease-broker/mutator-gate python unitests, `check-runtime-launches.py`, and ~30 framework-tool shell suites (git wrappers, wake, woodpecker, tmux, glpi, orchestrator, `_scripts`). Quality-relevant members rowed separately below                                                                                                                                                                                                       | `turbo test` → `CI ci.yml#test`; verify-release `test` stage                                                                                                                                                                                                                                                                                                                | QC-5 Test suite execution                                                                                                               | preserve                         | The chain is the execution surface the enumeration guard audits; known residuals: a failing link stops later suites (measured in #1270 — suites after position 44 had not run), and the guard proves naming, not reachability.                                    |
+
+### 1.3 Framework runtime hooks and their harnesses (agent-host enforcement)
+
+| check                                 | location                                                                                  | kind  | what it actually verifies                                                                                                                                                                                                                                                                                                                                                            | enforcement point                                                                | canonical check                         | disposition         | rationale                                                                                                                                                                                                  |
+| ------------------------------------- | ----------------------------------------------------------------------------------------- | ----- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------- | --------------------------------------- | ------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| QA edit hook seam                     | `framework/tools/qa/qa-hook-stdin.sh` (+ `qa-hook-handler.sh`)                            | shell | PostToolUse stdin hook: extracts edited file from the tool JSON (jq or grep fallback), skips non-JS/TS, then the deps-preflight gate — exits 1 with the legible sentinel `deps not installed — run pnpm install` when `node_modules/.bin` is missing/empty (the #856 false-red class); the downstream handler only files QA remediation **report templates** (no verification logic) | `agent-runtime` (framework `runtime/claude/settings.json` PostToolUse); never CI | QC-16 Agent-runtime edit-time checks    | strengthen (review) | The sentinel gate is real enforcement; the handler's report-filing adds no verdict and its name promises more than the code does — evaluator absorption should keep the sentinel, drop the report theater. |
+| Typecheck-on-edit hook                | `framework/tools/qa/typecheck-hook.sh`                                                    | shell | PostToolUse: for edited `.ts/.tsx`, finds nearest `tsconfig.json` and runs `tsc --noEmit`, surfacing errors nonzero to the agent immediately                                                                                                                                                                                                                                         | `agent-runtime` (framework `runtime/claude/settings.json` PostToolUse)           | QC-16 Agent-runtime edit-time checks    | strengthen (review) | Edit-time duplicate of QC-2 with independent invocation logic; keep behavior, converge invocation through the evaluator adapter.                                                                           |
+| Deps-preflight harness                | `framework/tools/qa/test-deps-preflight.sh`                                               | shell | Five assertions against the seam incl. a documented RED control (raw `not found`), sentinel behavior for missing and empty `.bin`, and no-false-positive once populated                                                                                                                                                                                                              | `test:framework-shell` → `CI ci.yml#test`                                        | QC-16 Agent-runtime edit-time checks    | preserve            | Guard-of-the-check with a red control; keeps the sentinel from regressing.                                                                                                                                 |
+| Prompt-helper RCE regression          | `framework/tools/_scripts/test-mosaic-init-rce.sh`                                        | shell | Sources the prompt helpers and proves a literal `$(touch /tmp/pwned)` answer round-trips verbatim and never executes (no `/tmp/pwned` created)                                                                                                                                                                                                                                       | `test:framework-shell` → `CI ci.yml#test`                                        | QC-5 Test suite execution               | preserve            | Cheap, load-bearing security regression on the installer's input path.                                                                                                                                     |
+| Install-ordering harness (#869 C2)    | `framework/tools/_scripts/test-install-ordering-guard.sh`                                 | shell | Drives `mosaic-link-runtime-assets` with a fake `mosaic` on PATH: probe ok ⇒ settings copied + exit 0; probe fail ⇒ exit 1 with degraded outcome but all other runtime files still copied; `--allow-inactive-enforcement` forwarded; no-mosaic-on-PATH ⇒ python3 fallback strips enforcement hooks and exits 1; fallback + flag ⇒ wires as-is, exit 0                                | `test:framework-shell` → `CI ci.yml#test`                                        | QC-17 Lease-enforcement wiring safety   | preserve            | Exercises the shell wiring seam independently of the TS guard's own spec suite (complementary coverage, by design).                                                                                        |
+| Fleet-transport harness (#1240)       | `framework/tools/_scripts/test-fleet-transport-check.sh`                                  | shell | Extracts the shipped `check_fleet_transport`/`fleet_declared_transport` functions **from the shipped scripts** (fails loud if extraction yields nothing) and drives both implementations (mosaic-doctor + `tools/install.sh`) from one case table                                                                                                                                    | `test:framework-shell` → `CI ci.yml#test`                                        | QC-18 Operator-host drift audit         | preserve            | The anti-drift harness for the one rule shipped twice; extraction-from-source keeps it from testing a stale copy.                                                                                          |
+| Terminal-green contract (RM-61/#1000) | `framework/tools/woodpecker/test-terminal-green-contract.sh` + `verify-terminal-green.py` | shell | Red-first fixtures: pipeline JSON variants (service failure, step failure, cancelled, etc.) must produce the correct terminal-green verdict; controls must pass                                                                                                                                                                                                                      | `test:framework-shell` → `CI ci.yml#test`                                        | QC-5 Test suite execution               | preserve            | Keeps the CI-wait wrapper's green-detection honest; a false green here would poison every merge gate that trusts `pr-ci-wait.sh`.                                                                          |
+| Lease-gate launch invariant           | `framework/tools/lease-broker/check-runtime-launches.py`                                  | shell | Scans production roots (`packages/`, `apps/`, `plugins/`, `tools/`) across sh/py/ts/yaml suffixes for Claude/Pi process launches **outside** the lease gate; allowlist-based; fails CI on violation                                                                                                                                                                                  | `test:framework-shell` → `CI ci.yml#test`                                        | QC-15 Lease-gate architecture invariant | preserve            | The only architectural "no ungated launches" rail; grep+allowlist is the right cost/benefit for this invariant.                                                                                            |
+
+### 1.4 TypeScript quality logic (`@mosaicstack/quality-rails` + mosaic CLI)
+
+| check                                    | location                                                                                       | kind     | what it actually verifies                                                                                                                                                                                                                                                                                                                                                                                                            | enforcement point                                                                                              | canonical check                       | disposition         | rationale                                                                                                                                                                                                                                                                           |
+| ---------------------------------------- | ---------------------------------------------------------------------------------------------- | -------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------- | ------------------------------------- | ------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `quality-rails check`                    | `packages/quality-rails/src/cli.ts` (`mosaic quality-rails check --project`)                   | ts       | **Expected-file presence only**: loops `expectedFilesForKind` (node: `.eslintrc`, `biome.json`, `.githooks/pre-commit`, `PR-CHECKLIST.md`; python: `pyproject.toml`+hooks+checklist; rust: `rustfmt.toml`+…) and exits 1 listing missing paths. Does not execute any linter, formatter, hook, or scanner                                                                                                                             | `local` (operator CLI); **no CI wiring in this repo**                                                          | QC-19 Downstream rails presence check | strengthen          | This is the RI-N4 evaluator seed. Today presence ≠ parity (explicitly called out by RI-N4): it must grow typed verdicts (`passed/failed/blocked/error/not-applicable`), check versioning/subject/reason, digested definitions, and absorb the effective shell probes (QC-20 first). |
+| `quality-rails doctor`                   | `packages/quality-rails/src/cli.ts`                                                            | ts       | Same presence data as `check`, printed with ok/missing lines; **cannot fail** (no nonzero exit on missing files)                                                                                                                                                                                                                                                                                                                     | `local` (operator CLI)                                                                                         | QC-19 Downstream rails presence check | strengthen          | A doctor that cannot fail is advisory; fold into `check` (or return typed states) when the evaluator lands.                                                                                                                                                                         |
+| `quality-rails init`                     | `packages/quality-rails/src/cli.ts` + `scaffolder.ts`/`templates.ts`                           | ts       | Scaffolds rails files per detected kind/profile (linters/formatters lists are advisory strings; hooks flag always true); writes files, prints follow-ups — no post-condition verification                                                                                                                                                                                                                                            | `local` (operator CLI)                                                                                         | QC-21 Downstream rails scaffolding    | strengthen (review) | Second scaffolding path alongside quality `install.sh` (§1.2); converge on one with post-scaffold verification before retiring either.                                                                                                                                              |
+| Lease activation probe (#869 C1, hidden) | `packages/mosaic/src/commands/lease-activation-probe.ts`                                       | ts       | Real capability probe, not file presence: resolves the installed mosaic CLI and requires it to advertise the exact `{name, version}` activation contract; all deps injectable; registered as hidden CLI command and consumed by C2/C5                                                                                                                                                                                                | `local` (hidden CLI + consumed by C2/C5); spec-tested via `lease-activation-probe.spec.ts` in `turbo test`     | QC-17 Lease-enforcement wiring safety | preserve            | The versioned-contract probe is precisely the fail-closed capability check RI-N2 generalizes; already typed and injectable.                                                                                                                                                         |
+| Install-ordering guard (#869 C2, hidden) | `packages/mosaic/src/commands/install-ordering-guard.ts`                                       | ts       | Decides whether enforcement hook entries are written into the `~/.claude/settings.json` the framework reseed ships: not activatable ⇒ strip hooks + nonzero loud outcome (default); explicit per-invocation `--allow-inactive-enforcement` opt-out wires-with-warning. Never touches the runtime gate's own fail-closed behavior                                                                                                     | `installer` (framework reseed via `mosaic-link-runtime-assets`); spec + shell harness coverage in `turbo test` | QC-17 Lease-enforcement wiring safety | preserve            | Correct default-deny with an explicit, non-env opt-out; test-locked from both the TS and shell sides.                                                                                                                                                                               |
+| Lease doctor check (#869 C5)             | `packages/mosaic/src/commands/lease-doctor-check.ts`                                           | ts       | Combines hook-wiring detection in `~/.claude/settings.json` with C1 activatable and C3 broker-supervisor health: wired ∧ (¬activatable ∨ ¬healthy) ⇒ loud `[ERROR]` that forces `mosaic doctor` exit 1 regardless of the bash audit's own exit                                                                                                                                                                                       | `local` (inside `mosaic doctor`); spec coverage in `turbo test`                                                | QC-17 Lease-enforcement wiring safety | preserve            | Closes the "bricked host looks green" hole; cannot be masked by the bash script — that composition is the point.                                                                                                                                                                    |
+| `mosaic doctor` (framework drift audit)  | `packages/mosaic/src/commands/launch.ts` (`doctor`) + `framework/tools/_scripts/mosaic-doctor` | shell+ts | Bash audit of the installed framework home: ~40 expected files/dirs present; runtime files are copies (not symlinks) matching source (`cmp`) or composed runtime-contract markers; hard-gates block present in AGENTS.md; sequential-thinking MCP configured; fleet transport binary present per roster (warn); legacy symlink trees gone; skills synced — **warn-based, exit 1 only with `--fail-on-warn`**, plus C5's forced error | `local` (operator audit)                                                                                       | QC-18 Operator-host drift audit       | preserve            | Host-state audit CI cannot see (user files by design, DESIGN §7); advisory exit is the documented contract — do not silently change it.                                                                                                                                             |
+| `mosaic gateway doctor`                  | `packages/mosaic/src/commands/gateway-doctor.ts`                                               | ts       | Probes per-service health (PostgreSQL, Valkey, pgvector) via `@mosaicstack/storage`, reports tier and JSON; exit 1 only when at least one **required** service fails (yellow stays 0)                                                                                                                                                                                                                                                | `local` (operator)                                                                                             | QC-18 Operator-host drift audit       | preserve            | Service health with correct red/yellow exit semantics; JSON mode exists for scripting.                                                                                                                                                                                              |
+| `mosaic gateway verify`                  | `packages/mosaic/src/commands/gateway/verify.ts`                                               | ts       | Post-install liveness: daemon meta via HTTP with retries, admin token on file, bootstrap endpoint reachable; aggregated pass/fail                                                                                                                                                                                                                                                                                                    | `local`; consumed by `tools/e2e-install-test.sh`                                                               | QC-18 Operator-host drift audit       | preserve            | The first-run proof the installer E2E relies on; retry-aware so startup races don't false-red.                                                                                                                                                                                      |
+| `mosaic fleet doctor`                    | `packages/mosaic/src/commands/fleet-reconciler-command.ts`                                     | ts       | Classifies local roster-owned drift (no mutation) from the parsed v2 roster                                                                                                                                                                                                                                                                                                                                                          | `local` (operator)                                                                                             | QC-18 Operator-host drift audit       | preserve            | Dry-run classification is the correct non-mutating audit shape.                                                                                                                                                                                                                     |
+
+### 1.5 Git hooks (developer machine)
+
+| check                     | location                                                  | kind  | what it actually verifies                                                                                                                                                                                                                                | enforcement point                                                                                       | canonical check             | disposition | rationale                                                                                                         |
+| ------------------------- | --------------------------------------------------------- | ----- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------- | --------------------------- | ----------- | ----------------------------------------------------------------------------------------------------------------- |
+| Pre-commit staged hygiene | `.husky/pre-commit` → `npx lint-staged` (`.lintstagedrc`) | shell | On staged files only: `prettier --write` + `eslint --fix` for ts/tsx/js/jsx; `prettier --write` for json/md/yaml/yml. **Mutating** (fixes and re-stages); commit blocks only if a fixer itself fails                                                     | `pre-commit` (every local commit; hooks activated by `install-hooks.mjs` via `core.hooksPath .husky/_`) | QC-13 Staged-change hygiene | preserve    | Correct scoped fast gate; note it auto-fixes rather than rejects (deliberate). Gap: no secret scan here — see §3. |
+| Pre-push gate             | `.husky/pre-push`                                         | shell | `pnpm preflight && pnpm typecheck && pnpm lint && pnpm format:check` (no test run — documented in AGENTS.md)                                                                                                                                             | `pre-push`                                                                                              | QC-14 Pre-push gate         | preserve    | Composes QC-1..4 exactly as specified in AGENTS.md; tests intentionally left to CI.                               |
+| Hook installer            | `scripts/install-hooks.mjs` (`pnpm prepare`)              | ts    | Stages husky hooks into a scratch repo first, asserts husky produced its `h` shim, quarantines incomplete previous sets, verifies idempotence via full directory snapshot comparison, then sets `core.hooksPath`; skips cleanly with `HUSKY=0` or no git | `installer` (runs on `pnpm install`)                                                                    | QC-13 Staged-change hygiene | preserve    | Self-verifying wiring for the hook gates — a corrupted half-install cannot silently disable them.                 |
+
+### 1.6 CI pipeline steps (`.woodpecker/`)
+
+Step-to-probe mapping for container steps: `ci.yml#sanitization` = QC-7+QC-8+QC-9 (rows §1.2, plus `apk add bash` env prep); `ci.yml#upgrade-guard` = QC-10 (rows §1.2, plus `apk add rsync`); `ci.yml#typecheck`/`#lint`/`#format`/`#test` = QC-2/3/4/5 (rows §1.1). Rows below are mechanisms that exist only in a pipeline.
+
+| check                                  | location                                                                               | kind          | what it actually verifies                                                                                                                                                                                                                                                                                                            | enforcement point                                    | canonical check                     | disposition                                                           | rationale                                                                                                     |
+| -------------------------------------- | -------------------------------------------------------------------------------------- | ------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ---------------------------------------------------- | ----------------------------------- | --------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------- | -------- | ------------------------------------------------------------------------------------------------------------- |
+| Frozen install                         | `ci.yml#install`                                                                       | pipeline-step | `pnpm install --frozen-lockfile --prefer-offline` against the baked ci-base store — lockfile supply integrity; a drifted lockfile fails the build before any gate runs                                                                                                                                                               | `CI ci.yml#install`                                  | QC-1 Checkout integrity             | preserve                                                              | Lockfile-pinned dep resolution is the supply-chain floor under every later gate.                              |
+| Test-step readiness prelude            | `ci.yml#test` prologue                                                                 | pipeline-step | Installs pinned `@earendil-works/[email protected]` (Invariant R suite requires the real binary) + openssl; waits up to 60×1s on `pg_isready` for the `ci-postgres` service and fails fast if it never comes up; runs `db:migrate` before tests                                                                                 | `CI ci.yml#test`                                     | QC-5 Test suite execution           | preserve                                                              | Fail-fast environment preconditions — a missing service produces a legible failure, not a wall of red tests.  |
+| Publish verify step (pending RI-1-001) | `publish.yml#verify` (branch `feat/ri-050-publish-gate` @ `46784c8d`, not yet on next) | pipeline-step | (a) Commit identity: fails closed if `CI_COMMIT_SHA` empty, `git rev-parse HEAD` empty, or the two differ; (b) runs the canonical `pnpm verify:release`. **Every publish effect depends on this step; it carries no path filter**                                                                                                    | `publish.yml#verify`                                 | QC-11 Terminal release verification | preserve                                                              | The RI-N1 exact-commit binding; until it merges, publish steps on next depend on `build` only (see §3 gap 1). |
+| Publish error classification           | `publish.yml#publish-npm`                                                              | pipeline-step | Publishes `@mosaicstack/*` (minus web) and classifies outcome: success, or the **only tolerated failure** = already-published (EPUBLISHCONFLICT / "cannot publish over" / "previously published"); explicit fatal on npm `E404/E401/ENEEDAUTH/ECONNREFUSED/ETIMEDOUT/ENOTFOUND` and on any unrecognized failure (replacing the old ` |                                                      | echo` that hid a registry 404)      | `publish.yml#publish-npm` (main/tags, path-filtered on `packages/**`) | QC-12 Publish-effect integrity                                                                                | preserve | Converts silent publish fall-on-floor into loud failure; allowlist-of-one error tolerance is the right shape. |
+| Next-lane publish assertions           | `publish.yml#publish-next-npm`                                                         | pipeline-step | Guards: branch must be `next`, `CI_PIPELINE_NUMBER` required; registry dist-tags JSON must be usable; walks all manifests, strictly parses stable semver, rewrites `X.Y.(Z+1)-next.<N>`; publishes with `--tag next` (never latest); post-publish asserts `npm view @mosaicstack/mosaic@next` resolves to the exact expected version | `publish.yml#publish-next-npm` (push/manual on next) | QC-12 Publish-effect integrity      | preserve                                                              | Durable prerelease lane with end-to-end resolution proof — the published artifact is verified, not assumed.   |
+| Image destination policy               | `publish.yml#build-gateway` / `#build-appservice` / `#build-web`                       | pipeline-step | Kaniko builds with destination policy: `next` ⇒ sha-tag only (fatal if a tag event sneaks in); `main` ⇒ sha + `latest`; tag events ⇒ sha + `<tag>`; anything else fatal. Path filters only skip **effects**, never the verify step                                                                                                   | `publish.yml#build-*`                                | QC-12 Publish-effect integrity      | preserve                                                              | Fail-closed tagging matrix; the exclude-list default-safe design keeps stale images impossible.               |
+
+Adjacent pipeline surface (not a probe): `.woodpecker/ci-image.yml` rebuilds the ci-base image on `pnpm-lock.yaml`/`Dockerfile.ci` change with an immutable `lock-<hash>` tag; pipelines consume `:latest`. Recorded for completeness — no code-quality property is checked.
+
+### 1.7 Root installer tooling (`tools/`)
+
+| check                       | location                                                  | kind  | what it actually verifies                                                                                                                                                                                                                                                              | enforcement point                                       | canonical check                 | disposition         | rationale                                                                                                                                                       |
+| --------------------------- | --------------------------------------------------------- | ----- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------- | ------------------------------- | ------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| Next-lane installer test    | `tools/install-next-lane.test.sh` (`pnpm test:installer`) | shell | Drives `tools/install.sh --next` with faked `node`/`npm` binaries (no network): Node 20 must be rejected; installs must pin **exact** versions (mutable `@next` forbidden); fast path must not unexpectedly fall back to source; gateway-install failure takes the documented fallback | `turbo`-external tail of `pnpm test` → `CI ci.yml#test` | QC-5 Test suite execution       | preserve            | Hermetic (shimmed) regression net for the installer lane; runs as part of the standard test command.                                                            |
+| Clean-container install E2E | `tools/e2e-install-test.sh`                               | shell | Full first-run flow in a node:22-alpine container: `install.sh --yes` → `mosaic wizard` (non-interactive) → `mosaic gateway install` → `mosaic gateway verify` exit check (with EXPECTED-SKIP if the installed CLI predates `gateway verify`); skips gracefully without Docker         | `local` (manual; requires Docker); **not wired in CI**  | QC-5 Test suite execution       | strengthen (review) | The only end-to-end proof of the install→verify path; currently operator-initiated only — wire into a periodic/manual CI lane or sign its exclusion explicitly. |
+| Host installer advisories   | `tools/install.sh` (`--check`; `check_fleet_transport`)   | shell | `--check` = version comparison only, no install; `check_fleet_transport` warns (non-blocking, by design — tmux is the fleet's dependency, not mosaic's) when the roster-declared transport binary is absent, naming exactly what it blocks; PATH-persistence warnings                  | `installer` (operator-run)                              | QC-18 Operator-host drift audit | preserve            | Advisory-by-design warnings; the parallel doctor check is drift-tested by §1.3's harness.                                                                       |
+
+### 1.8 Pending workstream additions (branch `feat/ri-050-publish-gate` @ `46784c8d`)
+
+| check                           | location                                             | kind | what it actually verifies                                                                                                                                                                                                                                                                          | enforcement point                                               | canonical check                     | disposition | rationale                                                                                      |
+| ------------------------------- | ---------------------------------------------------- | ---- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------- | ----------------------------------- | ----------- | ---------------------------------------------------------------------------------------------- |
+| Canonical terminal verification | `scripts/verify-release.mjs` (`pnpm verify:release`) | ts   | One command replaying the full mandatory set as stages — sanitization, upgrade-guard, typecheck (incl. preflight), lint, format, test, build — mirroring `ci.yml` step-for-step; fail-fast on first failing command; requires `bash`+`rsync` on PATH; `--stage <name>` for wiring smoke-tests only | `publish.yml#verify` (pending); `local` (`pnpm verify:release`) | QC-11 Terminal release verification | preserve    | The RI-N1 canonical command — CI and publication share one semantic checklist by construction. |
+| Verify-parity contract test     | `scripts/verify-release.test.mjs`                    | ts   | Parses the real `ci.yml`/`publish.yml`: stage table must match ci.yml step-for-step; every publish-effect step (name `publish*` or image-pushing) must transitively depend on `verify`; commit-identity assertion must be present; `verify` must carry no path filter                              | `test:checkout` → `CI ci.yml#test` (once merged)                | QC-11 Terminal release verification | preserve    | Guard-of-the-guard at checkout time — the two surfaces cannot drift apart silently.            |
+
+## 2. Canonical check set
+
+The deduplicated checks every row above maps onto. IDs are stable for RI-3-002 to consume.
+
+- **QC-1 Checkout integrity.** Owns: the checkout can run its gates — frozen-lockfile dependency resolution, required gate binaries present, no stale build lock, and the `apps/web/.next` generated-state trust chain (real directory, uid ownership, certified source fingerprint, certified symlink manifest). Implemented by `scripts/preflight.mjs` + frozen install steps.
+- **QC-2 Workspace typecheck.** Owns workspace-wide TypeScript soundness: per-package `tsc --noEmit` over built dependencies (`turbo typecheck`). The single definition invoked by CI, pre-push, and terminal verification.
+- **QC-3 Workspace lint.** Owns static-analysis policy: per-package ESLint under the root config. One config, one task, every surface.
+- **QC-4 Format check.** Owns formatting uniformity: Prettier check with the repo ignore list. (The pre-commit variant additionally fixes; the verdict form is this check.)
+- **QC-5 Test suite execution.** Owns execution of all test surfaces: checkout script units (`node --test`), per-package Vitest suites (including the framework shell chain and its python unitests), the installer-lane shim test, and — once wired — `test-roster-schema.py` and container E2E. Also owns guards-of-the-gate that live inside the chain (terminal-green contract, RCE regression).
+- **QC-6 Workspace build.** Owns artifact buildability: `turbo build` producing the artifacts publication consumes.
+- **QC-7 Framework sanitization.** Owns the open-source guarantee for the shipped framework package: no operator-identity tokens anywhere (examples included), no private `$HOME` defaults in shipped scripts, with a self-test that keeps the regexes honest.
+- **QC-8 Resident-context budget.** Owns the line-count ceilings on framework files injected into every agent's context (Constitution, dispatcher, RUNTIME.md slices) — the CI-enforceable half of the resident-prompt budget.
+- **QC-9 Test-membership enumeration.** Owns the property that no test suite can silently fall out of CI: disk population vs parsed enumeration surfaces, both-directions staleness, and signed exclusions with reasons. Includes its needle/control harness.
+- **QC-10 Upgrade/install safety.** Owns the #791 family: operator-path byte-identity across keep-mode upgrades (manifest guard), mid-failure rollback (errtrace-proven), durable pre-update snapshot + verify net + CWE-59 leaf guard, and the v2→v3 migration matrix with shell/TS parity.
+- **QC-11 Terminal release verification.** Owns the RI-N1 exact-commit binding: commit-identity assertion plus one canonical command (`pnpm verify:release`) replaying the complete mandatory set, with every publish effect depending on it; plus the checkout-time parity/DAG contract test that keeps pipeline and command in sync.
+- **QC-12 Publish-effect integrity.** Owns publication correctness: npm publish error classification (only already-published tolerated), next-lane versioning and post-publish resolution proof, and image destination/tag policy.
+- **QC-13 Staged-change hygiene.** Owns commit-time hygiene on staged files (prettier/eslint fix-and-restage) and the self-verifying hook wiring that guarantees the gates are actually installed.
+- **QC-14 Pre-push gate.** Owns the local push composition: preflight + typecheck + lint + format:check (tests deliberately deferred to CI).
+- **QC-15 Lease-gate architecture invariant.** Owns "no ungated runtime launches in production code": the scan + allowlist over `packages/`, `apps/`, `plugins/`, `tools/`.
+- **QC-16 Agent-runtime edit-time checks.** Owns edit-time feedback on agent hosts: the deps-preflight legibility sentinel and typecheck-on-edit, plus their regression harnesses.
+- **QC-17 Lease-enforcement wiring safety.** Owns the #869 C1/C2/C5 trio: activation capability probe (versioned contract), enforcement-hook wiring gate (default-deny with explicit opt-out), and the doctor check that surfaces a bricked host — with their shell/TS harnesses.
+- **QC-18 Operator-host drift audit.** Owns host-state health CI cannot see: `mosaic doctor` drift audit (+ fleet transport, both implementations), `fleet doctor` roster classification, `gateway doctor`/`gateway verify` service health, and installer advisories. Advisory exits are part of the contract.
+- **QC-19 Downstream rails presence check.** Owns "does a scaffolded project still carry its rails files" — today the TS `quality-rails check/doctor` presence loop; per RI-N4 this is the seed that must become the typed evaluator (presence alone is explicitly not parity).
+- **QC-20 Downstream enforcement verification.** Owns "do the rails actually block" on scaffolded projects: the behavioral planted-commit probe (type error, `any`, lint, gitleaks secret) currently in `verify.sh`/`verify.ps1` behind the `mosaic-quality-verify` adapter.
+- **QC-21 Downstream rails scaffolding.** Owns putting rails files into a target project: the shell template installer (+ PowerShell twin) and the TS `quality-rails init` scaffolder — currently two paths that must converge.
+
+## 3. Coverage gaps
+
+Enforced nowhere but implied, or named in docs/tooling but not wired:
+
+1. **Publication not yet bound to verification on `next`.** At this base (`8199261c`), `publish.yml` publish steps depend on `build` only; the `verify` step and `scripts/verify-release.mjs` exist on `feat/ri-050-publish-gate` (`46784c8d`) but are not merged. Until RI-1-001 lands, AC-RI-1's negative control cannot hold on the real pipeline.
+2. **Playwright E2E unwired.** `apps/web` ships `test:e2e` (`playwright test`) with real suites (`admin/auth/chat/navigation.spec.ts`); neither `pnpm test` nor any CI step invokes it. The web UI's user flows are verified only when an operator runs them manually.
+3. **No secret scanning on this repo.** The framework's own template pre-commit makes gitleaks **required**, and `verify.sh` proves detection with a planted key — but this repository's `.husky/pre-commit` (lint-staged only) and CI run no secret scan. The repo ships the control it does not use.
+4. **No dependency audit.** The quality `.woodpecker.yml` templates and `docs/CI-SETUP.md` specify `npm audit --audit-level=high` as a pipeline stage; nothing equivalent runs for this repo.
+5. **No coverage thresholds.** Templates enforce 80% Jest coverage thresholds; this repo's Vitest configs collect coverage with no thresholds — coverage is measured nowhere and enforced nowhere.
+6. **`test-roster-schema.py` invisible.** A real jsonschema regression suite wired to no surface and invisible to the enumeration guard (its population is `*.sh`; the suite is `.py`). Either enumerate it or sign an exclusion — silence here is the #1017 defect shape.
+7. **Presence-checker expectations ≠ this repo.** `quality-rails check` expects `.eslintrc`, `biome.json`, `.githooks/pre-commit`, `PR-CHECKLIST.md` for node projects — none describe this monorepo (husky, flat eslint config, no biome, no PR-CHECKLIST.md). The evaluator's check set must be per-subject (versioned, digested), not one global file list.
+8. **Chain-ordering residual (documented).** `test:framework-shell` is one `&&` chain: a failing link skips every later suite while the step still fails (measured in #1270 — four suites after position 44 had not run since a prior merge). The enumeration guard proves naming, not reachability; both residuals are in-file documented but structurally unfixed.
+9. **Signed-exclusion burndown open.** 16 signed exclusions remain in `test-enumeration-exclusions.txt`; several are "unmeasured in CI image" or blocked on missing CI tooling (tmux, setsid) — tracked under #1017/#1271. Each is an enforcement promise deferred, not delivered.
+10. **Windows twins unexercised.** `verify.ps1`, `install.ps1`, `mosaic-doctor.ps1` have no runner anywhere (no Windows CI); behavioral drift from their bash twins is undetectable by construction.
+11. **QA hook name vs behavior.** `qa-hook-handler.sh` files remediation report templates but performs no verification; the seam's actual gate value is only the deps-preflight sentinel. Anything relying on "QA automation hook" as a check is relying on report-filing.
+12. **Two test paths, one gated.** CI runs tests against ci-postgres (`DATABASE_URL` set); the local PGlite path is the documented default (AGENTS.md) until KBN-101-02/101-05. Only the CI path is enforced by pipeline.
+
+## 4. Disposition summary
+
+| disposition         | rows | checks                                                                                                                                                                                                                                                                                    |
+| ------------------- | ---- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| preserve            | 43   | Every canonical owner (QC-1..QC-18) plus correct guards-of-the-guard and thin adapters: all of §1.1, the CI-invoked framework probes and adapters in §1.2, all of §1.3, the C1/C2/C5 trio and doctors in §1.4, all of §1.5, all pipeline-only steps in §1.6, §1.7 rows 1 and 3, and §1.8. |
+| strengthen          | 2    | `quality-rails check` and `quality-rails doctor` (QC-19) — the RI-N4 evaluator seed: typed verdicts, versioned/digested check definitions, per-subject check sets.                                                                                                                        |
+| strengthen (review) | 9    | `verify.sh` + `verify.ps1` (QC-20), quality `install.sh`/`install.ps1` + `quality-rails init` (QC-21 — scaffold-path convergence), `test-roster-schema.py` (QC-5 — wire or sign), `qa-hook-stdin.sh` seam + `typecheck-hook.sh` (QC-16), `tools/e2e-install-test.sh` (QC-5 — CI lane).    |
+| retire              | 0    | None meet the bar: RI-N4 requires effective shell probes be **absorbed before** their paths retire, and no absorption exists yet. The `strengthen (review)` rows are the retirement candidates for RI-3-002 once the evaluator owns their behavior.                                       |
+
+Row total: 54. Canonical checks: 21 (QC-1..QC-21).

packages/macp/__tests__/gate-runner.test.ts

@@ -1,253 +0,0 @@
-import { mkdirSync, readFileSync, rmSync } from 'node:fs';
-import { join } from 'node:path';
-import { tmpdir } from 'node:os';
-import { randomUUID } from 'node:crypto';
-import { describe, it, expect, beforeEach, afterEach } from 'vitest';
-import { normalizeGate, countAIFindings, runGate, runGates } from '../src/gate-runner.js';
-
-function makeTmpDir(): string {
-  const dir = join(tmpdir(), `macp-gate-${randomUUID()}`);
-  mkdirSync(dir, { recursive: true });
-  return dir;
-}
-
-describe('normalizeGate', () => {
-  it('normalizes a string to mechanical gate', () => {
-    expect(normalizeGate('echo test')).toEqual({
-      command: 'echo test',
-      type: 'mechanical',
-      fail_on: 'blocker',
-    });
-  });
-
-  it('normalizes an object gate with defaults', () => {
-    expect(normalizeGate({ command: 'lint' })).toEqual({
-      command: 'lint',
-      type: 'mechanical',
-      fail_on: 'blocker',
-    });
-  });
-
-  it('preserves explicit type and fail_on', () => {
-    expect(normalizeGate({ command: 'review', type: 'ai-review', fail_on: 'any' })).toEqual({
-      command: 'review',
-      type: 'ai-review',
-      fail_on: 'any',
-    });
-  });
-
-  it('handles non-string/non-object input', () => {
-    expect(normalizeGate(42)).toEqual({ command: '', type: 'mechanical', fail_on: 'blocker' });
-    expect(normalizeGate(null)).toEqual({ command: '', type: 'mechanical', fail_on: 'blocker' });
-  });
-});
-
-describe('countAIFindings', () => {
-  it('returns zeros for non-object', () => {
-    expect(countAIFindings(null)).toEqual({ blockers: 0, total: 0 });
-    expect(countAIFindings('string')).toEqual({ blockers: 0, total: 0 });
-    expect(countAIFindings([])).toEqual({ blockers: 0, total: 0 });
-  });
-
-  it('counts from stats block', () => {
-    const output = { stats: { blockers: 2, should_fix: 3, suggestions: 1 } };
-    expect(countAIFindings(output)).toEqual({ blockers: 2, total: 6 });
-  });
-
-  it('counts from findings array when stats has no blockers', () => {
-    const output = {
-      stats: { blockers: 0 },
-      findings: [{ severity: 'blocker' }, { severity: 'warning' }, { severity: 'blocker' }],
-    };
-    expect(countAIFindings(output)).toEqual({ blockers: 2, total: 3 });
-  });
-
-  it('uses stats blockers over findings array when stats has blockers', () => {
-    const output = {
-      stats: { blockers: 5 },
-      findings: [{ severity: 'blocker' }, { severity: 'warning' }],
-    };
-    // stats.blockers = 5, total from stats = 5+0+0 = 5, findings not used for total since stats total is non-zero
-    expect(countAIFindings(output)).toEqual({ blockers: 5, total: 5 });
-  });
-
-  it('counts findings length as total when stats has zero total', () => {
-    const output = {
-      findings: [{ severity: 'warning' }, { severity: 'info' }],
-    };
-    expect(countAIFindings(output)).toEqual({ blockers: 0, total: 2 });
-  });
-});
-
-describe('runGate', () => {
-  let tmp: string;
-  let logPath: string;
-
-  beforeEach(() => {
-    tmp = makeTmpDir();
-    logPath = join(tmp, 'gate.log');
-  });
-
-  afterEach(() => {
-    rmSync(tmp, { recursive: true, force: true });
-  });
-
-  it('passes mechanical gate on exit 0', () => {
-    const result = runGate('echo hello', tmp, logPath, 30);
-    expect(result.passed).toBe(true);
-    expect(result.exit_code).toBe(0);
-    expect(result.type).toBe('mechanical');
-    expect(result.output).toContain('hello');
-  });
-
-  it('fails mechanical gate on non-zero exit', () => {
-    const result = runGate('exit 1', tmp, logPath, 30);
-    expect(result.passed).toBe(false);
-    expect(result.exit_code).toBe(1);
-  });
-
-  it('ci-pipeline always passes', () => {
-    const result = runGate({ command: 'anything', type: 'ci-pipeline' }, tmp, logPath, 30);
-    expect(result.passed).toBe(true);
-    expect(result.type).toBe('ci-pipeline');
-    expect(result.output).toBe('CI pipeline gate placeholder');
-  });
-
-  it('empty command passes', () => {
-    const result = runGate({ command: '' }, tmp, logPath, 30);
-    expect(result.passed).toBe(true);
-  });
-
-  it('ai-review gate parses JSON output', () => {
-    const json = JSON.stringify({ stats: { blockers: 0, should_fix: 1 } });
-    const result = runGate({ command: `echo '${json}'`, type: 'ai-review' }, tmp, logPath, 30);
-    expect(result.passed).toBe(true);
-    expect(result.blockers).toBe(0);
-    expect(result.findings).toBe(1);
-  });
-
-  it('ai-review gate fails on blockers', () => {
-    const json = JSON.stringify({ stats: { blockers: 2 } });
-    const result = runGate({ command: `echo '${json}'`, type: 'ai-review' }, tmp, logPath, 30);
-    expect(result.passed).toBe(false);
-    expect(result.blockers).toBe(2);
-  });
-
-  it('ai-review gate with fail_on=any fails on any findings', () => {
-    const json = JSON.stringify({ stats: { blockers: 0, should_fix: 1 } });
-    const result = runGate(
-      { command: `echo '${json}'`, type: 'ai-review', fail_on: 'any' },
-      tmp,
-      logPath,
-      30,
-    );
-    expect(result.passed).toBe(false);
-    expect(result.fail_on).toBe('any');
-  });
-
-  it('ai-review gate fails on invalid JSON output', () => {
-    const result = runGate({ command: 'echo "not json"', type: 'ai-review' }, tmp, logPath, 30);
-    expect(result.passed).toBe(false);
-    expect(result.parse_error).toBeDefined();
-  });
-
-  it('writes to log file', () => {
-    runGate('echo logged', tmp, logPath, 30);
-    const log = readFileSync(logPath, 'utf-8');
-    expect(log).toContain('COMMAND: echo logged');
-    expect(log).toContain('logged');
-    expect(log).toContain('EXIT:');
-  });
-});
-
-describe('runGates', () => {
-  let tmp: string;
-  let logPath: string;
-  let eventsPath: string;
-
-  beforeEach(() => {
-    tmp = makeTmpDir();
-    logPath = join(tmp, 'gates.log');
-    eventsPath = join(tmp, 'events.ndjson');
-  });
-
-  afterEach(() => {
-    rmSync(tmp, { recursive: true, force: true });
-  });
-
-  it('runs multiple gates and returns results', () => {
-    const { allPassed, gateResults } = runGates(
-      ['echo one', 'echo two'],
-      tmp,
-      logPath,
-      30,
-      eventsPath,
-      'task-1',
-    );
-    expect(allPassed).toBe(true);
-    expect(gateResults).toHaveLength(2);
-  });
-
-  it('reports failure when any gate fails', () => {
-    const { allPassed, gateResults } = runGates(
-      ['echo ok', 'exit 1'],
-      tmp,
-      logPath,
-      30,
-      eventsPath,
-      'task-2',
-    );
-    expect(allPassed).toBe(false);
-    expect(gateResults[0]!.passed).toBe(true);
-    expect(gateResults[1]!.passed).toBe(false);
-  });
-
-  it('emits events for each gate', () => {
-    runGates(['echo test'], tmp, logPath, 30, eventsPath, 'task-3');
-    const events = readFileSync(eventsPath, 'utf-8')
-      .trim()
-      .split('\n')
-      .map((l) => JSON.parse(l));
-    expect(events).toHaveLength(2); // started + passed
-    expect(events[0].event_type).toBe('rail.check.started');
-    expect(events[1].event_type).toBe('rail.check.passed');
-  });
-
-  it('skips gates with empty command (non ci-pipeline)', () => {
-    const { gateResults } = runGates(
-      [{ command: '', type: 'mechanical' }, 'echo real'],
-      tmp,
-      logPath,
-      30,
-      eventsPath,
-      'task-4',
-    );
-    expect(gateResults).toHaveLength(1);
-  });
-
-  it('does not skip ci-pipeline even with empty command', () => {
-    const { gateResults } = runGates(
-      [{ command: '', type: 'ci-pipeline' }],
-      tmp,
-      logPath,
-      30,
-      eventsPath,
-      'task-5',
-    );
-    expect(gateResults).toHaveLength(1);
-    expect(gateResults[0]!.passed).toBe(true);
-  });
-
-  it('emits failed event with correct message', () => {
-    runGates(['exit 42'], tmp, logPath, 30, eventsPath, 'task-6');
-    const events = readFileSync(eventsPath, 'utf-8')
-      .trim()
-      .split('\n')
-      .map((l) => JSON.parse(l));
-    const failEvent = events.find(
-      (e: Record<string, unknown>) => e.event_type === 'rail.check.failed',
-    );
-    expect(failEvent).toBeDefined();
-    expect(failEvent.message).toContain('Gate failed (');
-  });
-});

packages/macp/src/cli.spec.ts

@@ -1,5 +1,8 @@
-import { describe, it, expect } from 'vitest';
+import { describe, it, expect, afterEach, beforeEach, vi } from 'vitest';
 import { Command } from 'commander';
+import fs from 'node:fs';
+import os from 'node:os';
+import path from 'node:path';
 import { registerMacpCommand } from './cli.js';
 
 describe('registerMacpCommand', () => {
@@ -75,3 +78,162 @@ describe('registerMacpCommand', () => {
     expect(topLevel).toContain('events');
   });
 });
+
+/**
+ * RI-N2 fail-closed CLI behavior: an unimplemented capability is a failure,
+ * never a success. Every stub exits nonzero with a typed message, and the
+ * implemented `macp gate` mirrors the typed gate-runner states.
+ */
+describe('registerMacpCommand fail-closed (RI-N2)', () => {
+  let tmpDir: string;
+
+  function buildProgram(): Command {
+    const program = new Command();
+    program.exitOverride();
+    program.configureOutput({ writeErr: () => {} });
+    registerMacpCommand(program);
+    return program;
+  }
+
+  beforeEach(() => {
+    tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'macp-cli-failclosed-'));
+    process.exitCode = 0;
+  });
+
+  afterEach(() => {
+    process.exitCode = 0;
+    fs.rmSync(tmpDir, { recursive: true, force: true });
+  });
+
+  it('macp tasks list exits nonzero (unimplemented capability)', async () => {
+    const program = buildProgram();
+    await program.parseAsync(['macp', 'tasks', 'list'], { from: 'user' });
+    expect(process.exitCode).not.toBe(0);
+  });
+
+  it('macp submit exits nonzero with a typed MACP_NOT_IMPLEMENTED message', async () => {
+    const program = buildProgram();
+    const errSpy = vi.spyOn(console, 'error').mockImplementation(() => {});
+    try {
+      await program.parseAsync(['macp', 'submit', 'spec.json'], { from: 'user' });
+      expect(process.exitCode).not.toBe(0);
+      const errText = errSpy.mock.calls.map((c) => String(c[0])).join('\n');
+      expect(errText).toContain('MACP_NOT_IMPLEMENTED');
+    } finally {
+      errSpy.mockRestore();
+    }
+  });
+
+  it('macp events tail exits nonzero (unimplemented capability)', async () => {
+    const program = buildProgram();
+    await program.parseAsync(['macp', 'events', 'tail'], { from: 'user' });
+    expect(process.exitCode).not.toBe(0);
+  });
+
+  it('macp gate runs a green inline command and exits 0', async () => {
+    const program = buildProgram();
+    await program.parseAsync(
+      [
+        'macp',
+        'gate',
+        'exit 0',
+        '--cwd',
+        tmpDir,
+        '--log',
+        path.join(tmpDir, 'g.log'),
+        '--timeout',
+        '10',
+      ],
+      { from: 'user' },
+    );
+    expect(process.exitCode).toBe(0);
+  });
+
+  it('macp gate exits nonzero on a failing command', async () => {
+    const program = buildProgram();
+    await program.parseAsync(
+      [
+        'macp',
+        'gate',
+        'exit 9',
+        '--cwd',
+        tmpDir,
+        '--log',
+        path.join(tmpDir, 'g.log'),
+        '--timeout',
+        '10',
+      ],
+      { from: 'user' },
+    );
+    expect(process.exitCode).not.toBe(0);
+  });
+
+  it('macp gate with an unimplemented ci-pipeline capability exits nonzero', async () => {
+    const program = buildProgram();
+    const specPath = path.join(tmpDir, 'gates.json');
+    fs.writeFileSync(specPath, JSON.stringify([{ type: 'ci-pipeline' }]));
+    await program.parseAsync(
+      [
+        'macp',
+        'gate',
+        specPath,
+        '--cwd',
+        tmpDir,
+        '--log',
+        path.join(tmpDir, 'g.log'),
+        '--timeout',
+        '10',
+      ],
+      { from: 'user' },
+    );
+    expect(process.exitCode).not.toBe(0);
+  });
+
+  it('macp gate --simulate completes (exit 0) but reports simulated results', async () => {
+    const program = buildProgram();
+    const logSpy = vi.spyOn(console, 'log').mockImplementation(() => {});
+    try {
+      await program.parseAsync(
+        [
+          'macp',
+          'gate',
+          'exit 0',
+          '--simulate',
+          '--cwd',
+          tmpDir,
+          '--log',
+          path.join(tmpDir, 'g.log'),
+          '--timeout',
+          '10',
+        ],
+        { from: 'user' },
+      );
+      // completes only because the caller explicitly asked to simulate
+      expect(process.exitCode).toBe(0);
+      const outText = logSpy.mock.calls.map((c) => String(c[0])).join('\n');
+      expect(outText).toContain('simulated');
+      expect(outText).toContain('SIMULATED');
+    } finally {
+      logSpy.mockRestore();
+    }
+  });
+
+  it('macp gate with an empty spec exits nonzero with a typed error', async () => {
+    const program = buildProgram();
+    await program.parseAsync(
+      [
+        'macp',
+        'gate',
+        '   ',
+        '--cwd',
+        tmpDir,
+        '--log',
+        path.join(tmpDir, 'g.log'),
+        '--timeout',
+        '10',
+      ],
+      { from: 'user' },
+    );
+    expect(process.exitCode).not.toBe(0);
+  });
+});

packages/macp/src/cli.ts

@@ -1,5 +1,73 @@
+import { existsSync, readFileSync } from 'node:fs';
+
 import type { Command } from 'commander';
 
+import { runGates } from './gate-runner.js';
+import { MACPCapabilityError, type MacpErrorCode } from './errors.js';
+
+/**
+ * Load gates from a spec: an existing file (JSON gates array, a JSON object
+ * with `quality_gates`, a JSON gate object, or one command per line) or an
+ * inline command string. Fails closed with a typed capability error when the
+ * spec contains no executable gate definition.
+ */
+function loadGateSpec(spec: string): unknown[] {
+  if (existsSync(spec)) {
+    const raw = readFileSync(spec, 'utf-8');
+    try {
+      const parsed = JSON.parse(raw) as unknown;
+      if (Array.isArray(parsed)) {
+        if (parsed.length === 0) {
+          throw new MACPCapabilityError(
+            'MACP_NO_COMMAND',
+            'gate-spec',
+            `gate spec file '${spec}' contains an empty gates array`,
+          );
+        }
+        return parsed;
+      }
+      if (typeof parsed === 'object' && parsed !== null) {
+        const obj = parsed as Record<string, unknown>;
+        if (Array.isArray(obj['quality_gates'])) {
+          return obj['quality_gates'];
+        }
+        return [parsed];
+      }
+      throw new MACPCapabilityError(
+        'MACP_NO_COMMAND',
+        'gate-spec',
+        `gate spec file '${spec}' parsed to ${typeof parsed} — expected a gates array, a task with quality_gates, or a gate object`,
+      );
+    } catch (exc) {
+      if (exc instanceof MACPCapabilityError) throw exc;
+      // Not JSON — treat each non-empty line as a command gate.
+      const lines = raw
+        .split('\n')
+        .map((l) => l.trim())
+        .filter((l) => l.length > 0);
+      if (lines.length > 0) return lines;
+      throw new MACPCapabilityError(
+        'MACP_NO_COMMAND',
+        'gate-spec',
+        `gate spec file '${spec}' contains no gates`,
+      );
+    }
+  }
+  if (spec.trim().length > 0) return [spec];
+  throw new MACPCapabilityError('MACP_NO_COMMAND', 'gate-spec', 'gate spec is empty');
+}
+
+/** Print a typed not-implemented failure and exit nonzero (RI-N2 fail-closed). */
+function notImplemented(subcommand: string, capability: string, hint: string): void {
+  const err = new MACPCapabilityError(
+    'MACP_NOT_IMPLEMENTED',
+    capability,
+    `${subcommand} is not implemented in @mosaicstack/macp yet (${capability} capability absent) — ${hint}`,
+  );
+  console.error(`[macp] ${subcommand}: ${err.message} [${err.code}]`);
+  process.exitCode = 1;
+}
+
 /**
  * Register macp subcommands on an existing Commander program.
  * This avoids cross-package Commander version mismatches by using the
@@ -24,15 +92,14 @@ export function registerMacpCommand(parent: Command): void {
       'Filter by task type (coding|deploy|research|review|documentation|infrastructure)',
     )
     .action((opts: { status?: string; type?: string }) => {
-      // not yet wired — task persistence layer is not present in @mosaicstack/macp
+      // unimplemented capability — a failure, never a success (RI-N2)
-      console.log('[macp] tasks list: not yet wired — use macp package programmatically');
       if (opts.status) {
         console.log(`  status filter: ${opts.status}`);
       }
       if (opts.type) {
         console.log(`  type filter: ${opts.type}`);
       }
-      process.exitCode = 0;
+      notImplemented('tasks list', 'task-persistence', 'use the macp package programmatically');
     });
 
   // ─── submit ──────────────────────────────────────────────────────────────
@@ -41,12 +108,11 @@ export function registerMacpCommand(parent: Command): void {
     .command('submit <path>')
     .description('Submit a task from a JSON/YAML spec file')
     .action((specPath: string) => {
-      // not yet wired — task submission requires a running MACP server
+      // unimplemented capability — a failure, never a success (RI-N2)
-      console.log('[macp] submit: not yet wired — use macp package programmatically');
       console.log(`  spec path: ${specPath}`);
       console.log('  task id:   (unavailable — no MACP server connected)');
       console.log('  status:    (unavailable — no MACP server connected)');
-      process.exitCode = 0;
+      notImplemented('submit', 'macp-server', 'use the macp package programmatically');
     });
 
   // ─── gate ────────────────────────────────────────────────────────────────
@@ -58,16 +124,58 @@ export function registerMacpCommand(parent: Command): void {
     .option('--cwd <path>', 'Working directory for gate execution', process.cwd())
     .option('--log <path>', 'Path to write gate log output', '/tmp/macp-gate.log')
     .option('--timeout <seconds>', 'Gate timeout in seconds', '60')
-    .action((spec: string, opts: { failOn: string; cwd: string; log: string; timeout: string }) => {
+    .option(
-      // not yet wired — gate execution requires a task context and event sink
+      '--simulate',
-      console.log('[macp] gate: not yet wired — use macp package programmatically');
+      'Simulate gates instead of executing them; results are typed simulated and never satisfy a check',
-      console.log(`  spec:     ${spec}`);
+    )
-      console.log(`  fail-on:  ${opts.failOn}`);
+    .action(
-      console.log(`  cwd:      ${opts.cwd}`);
+      (
-      console.log(`  log:      ${opts.log}`);
+        spec: string,
-      console.log(`  timeout:  ${opts.timeout}s`);
+        opts: { failOn: string; cwd: string; log: string; timeout: string; simulate?: boolean },
-      process.exitCode = 0;
+      ) => {
-    });
+        let gates: unknown[];
+        try {
+          gates = loadGateSpec(spec);
+        } catch (exc) {
+          if (exc instanceof MACPCapabilityError) {
+            console.error(`[macp] gate: ${exc.message} [${exc.code}]`);
+          } else {
+            console.error(`[macp] gate: ${String(exc)}`);
+          }
+          process.exitCode = 1;
+          return;
+        }
+
+        const timeoutSec = Number.parseInt(opts.timeout, 10) || 60;
+        const eventsPath = `${opts.log}.events.ndjson`;
+        const { state, gateResults } = runGates(
+          gates,
+          opts.cwd,
+          opts.log,
+          timeoutSec,
+          eventsPath,
+          'macp-cli-gate',
+          {
+            simulate: opts.simulate,
+          },
+        );
+
+        for (const r of gateResults) {
+          const label = r.command || r.type;
+          const reason = r.reason ? ` — ${r.reason}` : '';
+          console.log(`[macp] gate ${r.status}: ${label}${reason}`);
+        }
+        if (opts.simulate) {
+          console.log(
+            '[macp] SIMULATED run — every result is typed simulated and can never satisfy a gate, dependency, or release check',
+          );
+        }
+
+        // Simulated runs may complete (exit 0) only because the caller
+        // explicitly passed --simulate; the typed state stays 'simulated'.
+        process.exitCode = state === 'passed' || state === 'simulated' ? 0 : 1;
+      },
+    );
 
   // ─── events ──────────────────────────────────────────────────────────────
 
@@ -79,14 +187,16 @@ export function registerMacpCommand(parent: Command): void {
     .option('--file <path>', 'Path to the MACP events NDJSON file')
     .option('--follow', 'Follow the file for new events (like tail -f)')
     .action((opts: { file?: string; follow?: boolean }) => {
-      // not yet wired — event streaming requires a live event source
+      // unimplemented capability — a failure, never a success (RI-N2)
-      console.log('[macp] events tail: not yet wired — use macp package programmatically');
       if (opts.file) {
         console.log(`  file:   ${opts.file}`);
       }
       if (opts.follow) {
         console.log('  mode:   follow');
       }
-      process.exitCode = 0;
+      notImplemented('events tail', 'event-source', 'use the macp package programmatically');
     });
 }
+
+// Re-export so CLI consumers can surface typed capability codes.
+export type { MacpErrorCode };

packages/macp/src/errors.ts

@@ -0,0 +1,35 @@
+/** Typed error code from the closed MACP_ERROR_CODES set. */
+export type MacpErrorCode = (typeof MACP_ERROR_CODES)[number];
+/**
+ * Typed fail-closed capability errors (RI-N2, SDLC-D-035).
+ *
+ * MACP must fail closed when a required capability (executor, reviewer,
+ * command, CI provider, human authority) is absent. These typed codes mirror
+ * the Forge failure vocabulary (FORGE_NO_*) so both packages speak the same
+ * language: an unimplemented capability is a failure, never a stub success.
+ */
+
+/** Closed set of typed MACP capability error codes. */
+export const MACP_ERROR_CODES = [
+  'MACP_NOT_IMPLEMENTED',
+  'MACP_NO_COMMAND',
+  'MACP_NO_REVIEWER',
+  'MACP_NO_CI_PIPELINE',
+  'MACP_NO_PROVIDER',
+  'MACP_AUTHORITY_REQUIRED',
+] as const;
+
+/** Raised when a required capability is missing and execution must fail closed. */
+export class MACPCapabilityError extends Error {
+  /** Typed error code from the closed MACP_ERROR_CODES set. */
+  readonly code: MacpErrorCode;
+  /** The missing capability, e.g. `ci-provider`, `task-persistence`, `command`. */
+  readonly capability: string;
+
+  constructor(code: MacpErrorCode, capability: string, message: string) {
+    super(message);
+    this.name = 'MACPCapabilityError';
+    this.code = code;
+    this.capability = capability;
+  }
+}

packages/macp/src/gate-runner.spec.ts

@@ -0,0 +1,429 @@
+import fs from 'node:fs';
+import os from 'node:os';
+import path from 'node:path';
+import { afterEach, beforeEach, describe, expect, it } from 'vitest';
+
+import { countAIFindings, normalizeGate, runGate, runGates } from './gate-runner.js';
+
+function makeTmpDir(): string {
+  return fs.mkdtempSync(path.join(os.tmpdir(), 'macp-gate-'));
+}
+
+describe('normalizeGate', () => {
+  it('normalizes a string to mechanical gate', () => {
+    expect(normalizeGate('echo test')).toEqual({
+      command: 'echo test',
+      type: 'mechanical',
+      fail_on: 'blocker',
+    });
+  });
+
+  it('normalizes an object gate with defaults', () => {
+    expect(normalizeGate({ command: 'lint' })).toEqual({
+      command: 'lint',
+      type: 'mechanical',
+      fail_on: 'blocker',
+    });
+  });
+
+  it('preserves explicit type and fail_on', () => {
+    expect(normalizeGate({ command: 'review', type: 'ai-review', fail_on: 'any' })).toEqual({
+      command: 'review',
+      type: 'ai-review',
+      fail_on: 'any',
+    });
+  });
+
+  it('handles non-string/non-object input', () => {
+    expect(normalizeGate(42)).toEqual({ command: '', type: 'mechanical', fail_on: 'blocker' });
+    expect(normalizeGate(null)).toEqual({ command: '', type: 'mechanical', fail_on: 'blocker' });
+  });
+});
+
+describe('countAIFindings', () => {
+  it('returns zeros for non-object', () => {
+    expect(countAIFindings(null)).toEqual({ blockers: 0, total: 0 });
+    expect(countAIFindings('string')).toEqual({ blockers: 0, total: 0 });
+    expect(countAIFindings([])).toEqual({ blockers: 0, total: 0 });
+  });
+
+  it('counts from stats block', () => {
+    const output = { stats: { blockers: 2, should_fix: 3, suggestions: 1 } };
+    expect(countAIFindings(output)).toEqual({ blockers: 2, total: 6 });
+  });
+
+  it('counts from findings array when stats has no blockers', () => {
+    const output = {
+      stats: { blockers: 0 },
+      findings: [{ severity: 'blocker' }, { severity: 'warning' }, { severity: 'blocker' }],
+    };
+    expect(countAIFindings(output)).toEqual({ blockers: 2, total: 3 });
+  });
+
+  it('uses stats blockers over findings array when stats has blockers', () => {
+    const output = {
+      stats: { blockers: 5 },
+      findings: [{ severity: 'blocker' }, { severity: 'warning' }],
+    };
+    // stats.blockers = 5, total from stats = 5+0+0 = 5, findings not used for total since stats total is non-zero
+    expect(countAIFindings(output)).toEqual({ blockers: 5, total: 5 });
+  });
+
+  it('counts findings length as total when stats has zero total', () => {
+    const output = {
+      findings: [{ severity: 'warning' }, { severity: 'info' }],
+    };
+    expect(countAIFindings(output)).toEqual({ blockers: 0, total: 2 });
+  });
+});
+
+describe('runGate', () => {
+  let tmp: string;
+  let logPath: string;
+
+  beforeEach(() => {
+    tmp = makeTmpDir();
+    logPath = path.join(tmp, 'gate.log');
+  });
+
+  afterEach(() => {
+    fs.rmSync(tmp, { recursive: true, force: true });
+  });
+
+  it('passes mechanical gate on exit 0', () => {
+    const result = runGate('echo hello', tmp, logPath, 30);
+    expect(result.passed).toBe(true);
+    expect(result.exit_code).toBe(0);
+    expect(result.type).toBe('mechanical');
+    expect(result.output).toContain('hello');
+  });
+
+  it('fails mechanical gate on non-zero exit', () => {
+    const result = runGate('exit 1', tmp, logPath, 30);
+    expect(result.passed).toBe(false);
+    expect(result.exit_code).toBe(1);
+  });
+
+  it('ci-pipeline fails closed without a CI provider (no placeholder pass)', () => {
+    const result = runGate({ command: 'anything', type: 'ci-pipeline' }, tmp, logPath, 30);
+    expect(result.passed).toBe(false);
+    expect(result.status).toBe('capability_failure');
+    expect(result.capability_code).toBe('MACP_NO_CI_PIPELINE');
+    expect(result.type).toBe('ci-pipeline');
+    expect(result.output).not.toBe('CI pipeline gate placeholder');
+  });
+
+  it('empty command is a typed capability failure, never a pass', () => {
+    const result = runGate({ command: '' }, tmp, logPath, 30);
+    expect(result.passed).toBe(false);
+    expect(result.status).toBe('capability_failure');
+    expect(result.capability_code).toBe('MACP_NO_COMMAND');
+  });
+
+  it('ai-review gate parses JSON output', () => {
+    const json = JSON.stringify({ stats: { blockers: 0, should_fix: 1 } });
+    const result = runGate({ command: `echo '${json}'`, type: 'ai-review' }, tmp, logPath, 30);
+    expect(result.passed).toBe(true);
+    expect(result.blockers).toBe(0);
+    expect(result.findings).toBe(1);
+  });
+
+  it('ai-review gate fails on blockers', () => {
+    const json = JSON.stringify({ stats: { blockers: 2 } });
+    const result = runGate({ command: `echo '${json}'`, type: 'ai-review' }, tmp, logPath, 30);
+    expect(result.passed).toBe(false);
+    expect(result.blockers).toBe(2);
+  });
+
+  it('ai-review gate with fail_on=any fails on any findings', () => {
+    const json = JSON.stringify({ stats: { blockers: 0, should_fix: 1 } });
+    const result = runGate(
+      { command: `echo '${json}'`, type: 'ai-review', fail_on: 'any' },
+      tmp,
+      logPath,
+      30,
+    );
+    expect(result.passed).toBe(false);
+    expect(result.fail_on).toBe('any');
+  });
+
+  it('ai-review gate fails on invalid JSON output', () => {
+    const result = runGate({ command: 'echo "not json"', type: 'ai-review' }, tmp, logPath, 30);
+    expect(result.passed).toBe(false);
+    expect(result.parse_error).toBeDefined();
+  });
+
+  it('writes to log file', () => {
+    runGate('echo logged', tmp, logPath, 30);
+    const log = fs.readFileSync(logPath, 'utf-8');
+    expect(log).toContain('COMMAND: echo logged');
+    expect(log).toContain('logged');
+    expect(log).toContain('EXIT:');
+  });
+});
+
+describe('runGates', () => {
+  let tmp: string;
+  let logPath: string;
+  let eventsPath: string;
+
+  beforeEach(() => {
+    tmp = makeTmpDir();
+    logPath = path.join(tmp, 'gates.log');
+    eventsPath = path.join(tmp, 'events.ndjson');
+  });
+
+  afterEach(() => {
+    fs.rmSync(tmp, { recursive: true, force: true });
+  });
+
+  it('runs multiple gates and returns results', () => {
+    const { allPassed, gateResults } = runGates(
+      ['echo one', 'echo two'],
+      tmp,
+      logPath,
+      30,
+      eventsPath,
+      'task-1',
+    );
+    expect(allPassed).toBe(true);
+    expect(gateResults).toHaveLength(2);
+  });
+
+  it('reports failure when any gate fails', () => {
+    const { allPassed, gateResults } = runGates(
+      ['echo ok', 'exit 1'],
+      tmp,
+      logPath,
+      30,
+      eventsPath,
+      'task-2',
+    );
+    expect(allPassed).toBe(false);
+    expect(gateResults[0]!.passed).toBe(true);
+    expect(gateResults[1]!.passed).toBe(false);
+  });
+
+  it('emits events for each gate', () => {
+    runGates(['echo test'], tmp, logPath, 30, eventsPath, 'task-3');
+    const events = fs
+      .readFileSync(eventsPath, 'utf-8')
+      .trim()
+      .split('\n')
+      .map((l) => JSON.parse(l));
+    expect(events).toHaveLength(2); // started + passed
+    expect(events[0].event_type).toBe('rail.check.started');
+    expect(events[1].event_type).toBe('rail.check.passed');
+  });
+
+  it('does not silently skip gates with empty command — they become capability failures', () => {
+    const { gateResults, allPassed, state } = runGates(
+      [{ command: '', type: 'mechanical' }, 'echo real'],
+      tmp,
+      logPath,
+      30,
+      eventsPath,
+      'task-4',
+    );
+    expect(gateResults).toHaveLength(2);
+    expect(gateResults[0]!.status).toBe('capability_failure');
+    expect(gateResults[1]!.status).toBe('passed');
+    expect(allPassed).toBe(false);
+    expect(state).toBe('capability_failure');
+  });
+
+  it('does not skip ci-pipeline even with empty command — typed capability failure', () => {
+    const { gateResults, allPassed, state } = runGates(
+      [{ command: '', type: 'ci-pipeline' }],
+      tmp,
+      logPath,
+      30,
+      eventsPath,
+      'task-5',
+    );
+    expect(gateResults).toHaveLength(1);
+    expect(gateResults[0]!.passed).toBe(false);
+    expect(gateResults[0]!.status).toBe('capability_failure');
+    expect(allPassed).toBe(false);
+    expect(state).toBe('capability_failure');
+  });
+
+  it('emits failed event with correct message', () => {
+    runGates(['exit 42'], tmp, logPath, 30, eventsPath, 'task-6');
+    const events = fs
+      .readFileSync(eventsPath, 'utf-8')
+      .trim()
+      .split('\n')
+      .map((l) => JSON.parse(l));
+    const failEvent = events.find(
+      (e: Record<string, unknown>) => e.event_type === 'rail.check.failed',
+    );
+    expect(failEvent).toBeDefined();
+    expect(failEvent.message).toContain('Gate failed (');
+  });
+});
+
+/**
+ * RI-N2 / SDLC-D-035 fail-closed controls for the MACP gate runner.
+ *
+ * Invariant under test: `passed: true` occurs ONLY when a gate really executed
+ * and really exited green (`status === 'passed'`). Absent capabilities,
+ * manual sign-offs, and simulated runs are typed distinctly and can never
+ * make the aggregate `passed`.
+ */
+describe('gate-runner fail-closed (RI-N2)', () => {
+  let tmpDir: string;
+  let logPath: string;
+  let eventsPath: string;
+
+  beforeEach(() => {
+    tmpDir = makeTmpDir();
+    logPath = path.join(tmpDir, 'gate.log');
+    eventsPath = path.join(tmpDir, 'events.ndjson');
+  });
+
+  afterEach(() => {
+    fs.rmSync(tmpDir, { recursive: true, force: true });
+  });
+
+  function run(gates: unknown[], options?: { simulate?: boolean }) {
+    return runGates(gates, tmpDir, logPath, 10, eventsPath, 'spec-task', options);
+  }
+
+  // ─── positive controls ───────────────────────────────────────────────────
+
+  it('a really-executed green command gate still passes', () => {
+    const result = run([{ command: 'exit 0', type: 'mechanical' }]);
+    expect(result.gateResults[0]!.status).toBe('passed');
+    expect(result.gateResults[0]!.passed).toBe(true);
+    expect(result.allPassed).toBe(true);
+    expect(result.state).toBe('passed');
+  });
+
+  it('explicit simulate completes and types every result simulated', () => {
+    const result = run([{ command: 'exit 0', type: 'mechanical' }, 'echo hello'], {
+      simulate: true,
+    });
+    expect(result.gateResults).toHaveLength(2);
+    for (const gate of result.gateResults) {
+      expect(gate.status).toBe('simulated');
+      expect(gate.passed).toBe(false);
+    }
+    expect(result.state).toBe('simulated');
+  });
+
+  it('a really-executed red command gate fails with typed status failed', () => {
+    const result = run([{ command: 'exit 3', type: 'mechanical' }]);
+    expect(result.gateResults[0]!.status).toBe('failed');
+    expect(result.gateResults[0]!.passed).toBe(false);
+    expect(result.allPassed).toBe(false);
+    expect(result.state).toBe('failed');
+  });
+
+  // ─── negative controls — each asserts typed status AND aggregate not passed ──
+
+  it('an empty-command gate is a capability_failure, not skipped and not passed', () => {
+    const result = run([{ command: '', type: 'mechanical' }]);
+    // runGates must not silently skip it — it produces a typed result
+    expect(result.gateResults).toHaveLength(1);
+    const gate = result.gateResults[0]!;
+    expect(gate.status).toBe('capability_failure');
+    expect(gate.capability_code).toBe('MACP_NO_COMMAND');
+    expect(gate.passed).toBe(false);
+    // aggregate is not passed
+    expect(result.allPassed).toBe(false);
+    expect(result.state).toBe('capability_failure');
+    expect(result.state).not.toBe('passed');
+  });
+
+  it('a commandless ai-review gate is a typed MACP_NO_REVIEWER capability_failure', () => {
+    const result = run([{ command: '', type: 'ai-review' }]);
+    expect(result.gateResults[0]!.status).toBe('capability_failure');
+    expect(result.gateResults[0]!.capability_code).toBe('MACP_NO_REVIEWER');
+    expect(result.allPassed).toBe(false);
+    expect(result.state).not.toBe('passed');
+  });
+
+  it('a ci-pipeline gate without a provider implementation is a capability_failure, never a placeholder pass', () => {
+    const result = run([{ command: '', type: 'ci-pipeline' }]);
+    const gate = result.gateResults[0]!;
+    expect(gate.status).toBe('capability_failure');
+    expect(gate.capability_code).toBe('MACP_NO_CI_PIPELINE');
+    expect(gate.passed).toBe(false);
+    // the old false-success placeholder must be gone
+    expect(gate.output).not.toBe('CI pipeline gate placeholder');
+    expect(result.allPassed).toBe(false);
+    expect(result.state).not.toBe('passed');
+  });
+
+  it('a ci-pipeline gate fails closed even alongside an otherwise green run', () => {
+    const result = run(['exit 0', { type: 'ci-pipeline', command: 'fake-ci' }]);
+    expect(result.gateResults[1]!.status).toBe('capability_failure');
+    expect(result.gateResults[0]!.status).toBe('passed');
+    expect(result.allPassed).toBe(false);
+    expect(result.state).toBe('capability_failure');
+  });
+
+  it('a manual gate with no automation enters typed waiting — neither pass nor fail', () => {
+    const result = run([{ type: 'manual' }]);
+    const gate = result.gateResults[0]!;
+    expect(gate.status).toBe('waiting');
+    expect(gate.passed).toBe(false);
+    expect(gate.exit_code).toBe(0);
+    // aggregate is not passed while any gate is waiting
+    expect(result.allPassed).toBe(false);
+    expect(result.state).toBe('waiting');
+    expect(result.state).not.toBe('passed');
+  });
+
+  it('a simulated result can never make the aggregate passed', () => {
+    const result = run(['exit 0', 'exit 0'], { simulate: true });
+    expect(result.gateResults.every((g) => g.status === 'simulated')).toBe(true);
+    expect(result.allPassed).toBe(false);
+    expect(result.state).toBe('simulated');
+    expect(result.state).not.toBe('passed');
+  });
+
+  it('waiting dominates an otherwise green aggregate', () => {
+    const result = run(['exit 0', { type: 'manual' }]);
+    expect(result.allPassed).toBe(false);
+    expect(result.state).toBe('waiting');
+  });
+});
+
+describe('runGate fail-closed (RI-N2)', () => {
+  let tmpDir: string;
+  let logPath: string;
+
+  beforeEach(() => {
+    tmpDir = makeTmpDir();
+    logPath = path.join(tmpDir, 'gate.log');
+  });
+
+  afterEach(() => {
+    fs.rmSync(tmpDir, { recursive: true, force: true });
+  });
+
+  it('simulate: true returns a typed simulated result without executing', () => {
+    const result = runGate('this-command-does-not-exist-xyz', tmpDir, logPath, 10, {
+      simulate: true,
+    });
+    expect(result.status).toBe('simulated');
+    expect(result.passed).toBe(false);
+    expect(result.exit_code).toBe(0);
+  });
+
+  it('normal mode executes for real and types a green gate passed', () => {
+    const result = runGate('echo ok', tmpDir, logPath, 10);
+    expect(result.status).toBe('passed');
+    expect(result.passed).toBe(true);
+    expect(result.output).toContain('ok');
+  });
+
+  it('a bare string gate normalizes to mechanical and executes', () => {
+    const result = runGate('exit 7', tmpDir, logPath, 10);
+    expect(result.type).toBe('mechanical');
+    expect(result.status).toBe('failed');
+    expect(result.passed).toBe(false);
+  });
+});

packages/macp/src/gate-runner.ts

@@ -4,7 +4,20 @@ import { dirname } from 'node:path';
 
 import { emitEvent } from './event-emitter.js';
 import { nowISO } from './event-emitter.js';
-import type { GateResult } from './types.js';
+import type { GateResult, GateStatus, RunGatesResult } from './types.js';
+
+/** Typed reason stamped on every simulated gate result. */
+export const SIMULATED_GATE_REASON =
+  'simulated execution (explicit simulate opt-in): gate was not evaluated by a real implementation';
+
+/** Options for gate execution (RI-N2 fail-closed / explicit simulation). */
+export interface RunGateOptions {
+  /**
+   * Explicit caller opt-in to simulation. Simulated gates are NOT executed;
+   * every result is typed `simulated` and never satisfies anything.
+   */
+  simulate?: boolean;
+}
 
 export interface NormalizedGate {
   command: string;
@@ -103,36 +116,91 @@ export function countAIFindings(parsedOutput: unknown): { blockers: number; tota
   return { blockers, total };
 }
 
+function simulatedResult(gateEntry: NormalizedGate): GateResult {
+  return {
+    command: gateEntry.command,
+    exit_code: 0,
+    type: gateEntry.type,
+    output: SIMULATED_GATE_REASON,
+    timed_out: false,
+    passed: false,
+    status: 'simulated',
+    reason: SIMULATED_GATE_REASON,
+  };
+}
+
+function capabilityFailureResult(
+  gateEntry: NormalizedGate,
+  code: GateResult['capability_code'],
+  reason: string,
+): GateResult {
+  return {
+    command: gateEntry.command,
+    exit_code: 1,
+    type: gateEntry.type,
+    output: '',
+    timed_out: false,
+    passed: false,
+    status: 'capability_failure',
+    capability_code: code,
+    reason,
+  };
+}
+
+function waitingResult(gateEntry: NormalizedGate, reason: string): GateResult {
+  return {
+    command: gateEntry.command,
+    exit_code: 0,
+    type: gateEntry.type,
+    output: '',
+    timed_out: false,
+    passed: false,
+    status: 'waiting',
+    capability_code: 'MACP_AUTHORITY_REQUIRED',
+    reason,
+  };
+}
+
 export function runGate(
   gate: unknown,
   cwd: string,
   logPath: string,
   timeoutSec: number,
+  options: RunGateOptions = {},
 ): GateResult {
   const gateEntry = normalizeGate(gate);
   const gateType = gateEntry.type;
   const command = gateEntry.command;
 
+  // Explicit simulation only: never executes, typed simulated, never satisfying.
+  if (options.simulate) {
+    return simulatedResult(gateEntry);
+  }
+
+  // Fail closed: no CI provider implementation exists in @mosaicstack/macp,
+  // so a ci-pipeline gate is an absent capability — never a placeholder pass.
   if (gateType === 'ci-pipeline') {
-    return {
+    return capabilityFailureResult(
-      command,
+      gateEntry,
-      exit_code: 0,
+      'MACP_NO_CI_PIPELINE',
-      type: gateType,
+      `ci-pipeline gate '${gateEntry.command || gateType}' has no CI provider implementation wired — refusing placeholder pass`,
-      output: 'CI pipeline gate placeholder',
+    );
-      timed_out: false,
-      passed: true,
-    };
   }
 
   if (!command) {
-    return {
+    // A manual gate with no automation waits for human sign-off: not pass, not fail.
-      command: '',
+    if (gateType === 'manual') {
-      exit_code: 0,
+      return waitingResult(
-      type: gateType,
+        gateEntry,
-      output: '',
+        `manual gate has no automation — waiting for human sign-off (type: ${gateType})`,
-      timed_out: false,
+      );
-      passed: true,
+    }
-    };
+    // Any other commandless gate is an absent capability — never a vacuous pass.
+    return capabilityFailureResult(
+      gateEntry,
+      gateType === 'ai-review' ? 'MACP_NO_REVIEWER' : 'MACP_NO_COMMAND',
+      `gate of type '${gateType}' has no command to execute — refusing empty-command pass`,
+    );
   }
 
   const { exitCode, output, timedOut } = runShell(command, cwd, logPath, timeoutSec);
@@ -143,10 +211,12 @@ export function runGate(
     output,
     timed_out: timedOut,
     passed: false,
+    status: 'failed',
   };
 
   if (gateType !== 'ai-review') {
     result.passed = exitCode === 0;
+    result.status = result.passed ? 'passed' : 'failed';
     return result;
   }
 
@@ -170,6 +240,7 @@ export function runGate(
   } else {
     result.passed = exitCode === 0 && blockers === 0 && !timedOut && parseError === undefined;
   }
+  result.status = result.passed ? 'passed' : 'failed';
 
   result.fail_on = failOn;
   result.blockers = blockers;
@@ -191,16 +262,19 @@ export function runGates(
   timeoutSec: number,
   eventsPath: string,
   taskId: string,
-): { allPassed: boolean; gateResults: GateResult[] } {
+  options: RunGateOptions = {},
-  let allPassed = true;
+): RunGatesResult {
   const gateResults: GateResult[] = [];
+  let hasCapabilityFailure = false;
+  let hasSimulated = false;
+  let hasFailed = false;
+  let hasWaiting = false;
 
   for (const gate of gates) {
     const gateEntry = normalizeGate(gate);
     const gateCmd = gateEntry.command;
-    if (!gateCmd && gateEntry.type !== 'ci-pipeline') continue;
-
     const label = gateCmd || gateEntry.type;
+    // NOTE: no silent skip — every gate produces a typed result (RI-N2).
     emitEvent(
       eventsPath,
       'rail.check.started',
@@ -209,10 +283,10 @@ export function runGates(
       'quality-gate',
       `Running gate: ${label}`,
     );
-    const result = runGate(gate, cwd, logPath, timeoutSec);
+    const result = runGate(gate, cwd, logPath, timeoutSec, options);
     gateResults.push(result);
 
-    if (result.passed) {
+    if (result.status === 'passed') {
       emitEvent(
         eventsPath,
         'rail.check.passed',
@@ -224,7 +298,46 @@ export function runGates(
       continue;
     }
 
-    allPassed = false;
+    if (result.status === 'waiting') {
+      hasWaiting = true;
+      emitEvent(
+        eventsPath,
+        'rail.check.waiting',
+        taskId,
+        'gated',
+        'quality-gate',
+        `Gate waiting: ${label} — ${result.reason ?? 'manual gate awaits sign-off'}`,
+      );
+      continue;
+    }
+
+    if (result.status === 'simulated') {
+      hasSimulated = true;
+      emitEvent(
+        eventsPath,
+        'rail.check.simulated',
+        taskId,
+        'gated',
+        'quality-gate',
+        `Gate simulated (non-satisfying): ${label}`,
+      );
+      continue;
+    }
+
+    if (result.status === 'capability_failure') {
+      hasCapabilityFailure = true;
+      emitEvent(
+        eventsPath,
+        'rail.check.failed',
+        taskId,
+        'gated',
+        'quality-gate',
+        `Gate capability failure (${result.capability_code ?? 'MACP_NO_PROVIDER'}): ${label} — ${result.reason ?? 'required capability is absent'}`,
+      );
+      continue;
+    }
+
+    hasFailed = true;
     let message: string;
     if (result.timed_out) {
       message = `Gate timed out after ${timeoutSec}s: ${label}`;
@@ -236,5 +349,15 @@ export function runGates(
     emitEvent(eventsPath, 'rail.check.failed', taskId, 'gated', 'quality-gate', message);
   }
 
-  return { allPassed, gateResults };
+  const state: GateStatus = hasCapabilityFailure
+    ? 'capability_failure'
+    : hasSimulated
+      ? 'simulated'
+      : hasFailed
+        ? 'failed'
+        : hasWaiting
+          ? 'waiting'
+          : 'passed';
+
+  return { allPassed: state === 'passed', gateResults, state };
 }

packages/macp/src/index.ts

@@ -6,11 +6,13 @@ export type {
   DependsOnPolicy,
   GateType,
   GateFailOn,
+  GateStatus,
   GateEntry,
   Task,
   EventType,
   MACPEvent,
   GateResult,
+  RunGatesResult,
   TaskResult,
   ProviderMeta,
   ProviderRegistry,
@@ -18,6 +20,11 @@ export type {
 
 export { CredentialError } from './types.js';
 
+// Typed fail-closed capability errors (RI-N2, SDLC-D-035)
+export { MACP_ERROR_CODES, MACPCapabilityError } from './errors.js';
+
+export type { MacpErrorCode } from './errors.js';
+
 // Credential resolver
 export {
   DEFAULT_CREDENTIALS_DIR,
@@ -35,9 +42,16 @@ export {
 export type { ResolveCredentialsOptions } from './credential-resolver.js';
 
 // Gate runner
-export { normalizeGate, runShell, countAIFindings, runGate, runGates } from './gate-runner.js';
+export {
+  normalizeGate,
+  runShell,
+  countAIFindings,
+  runGate,
+  runGates,
+  SIMULATED_GATE_REASON,
+} from './gate-runner.js';
 
-export type { NormalizedGate } from './gate-runner.js';
+export type { NormalizedGate, RunGateOptions } from './gate-runner.js';
 
 // Risk-floor (agent reflection loop — diff review classifier)
 export { evaluateRiskFloor, DEFAULT_RISK_THRESHOLD } from './risk-floor.js';

packages/macp/src/types.ts

@@ -1,3 +1,5 @@
+import type { MacpErrorCode } from './errors.js';
+
 /** Task status values. */
 export type TaskStatus = 'pending' | 'running' | 'gated' | 'completed' | 'failed' | 'escalated';
 
@@ -17,7 +19,17 @@ export type DispatchMode = 'yolo' | 'acp' | 'exec';
 export type DependsOnPolicy = 'all' | 'any' | 'all_terminal';
 
 /** Quality gate type. */
-export type GateType = 'mechanical' | 'ai-review' | 'ci-pipeline';
+export type GateType = 'mechanical' | 'ai-review' | 'ci-pipeline' | 'manual';
+
+/**
+ * Typed execution state of a gate — closed set (RI-N2, SDLC-D-035).
+ *
+ * Only `passed` means "really executed and green". `simulated` is produced
+ * exclusively under an explicit simulate opt-in and never satisfies anything.
+ * `capability_failure` means a required executor/provider/command was absent.
+ * `waiting` means a manual gate awaits human sign-off (neither pass nor fail).
+ */
+export type GateStatus = 'passed' | 'failed' | 'simulated' | 'waiting' | 'capability_failure';
 
 /** Gate fail_on mode. */
 export type GateFailOn = 'blocker' | 'any';
@@ -67,7 +79,9 @@ export type EventType =
   | 'task.retry.scheduled'
   | 'rail.check.started'
   | 'rail.check.passed'
-  | 'rail.check.failed';
+  | 'rail.check.failed'
+  | 'rail.check.waiting'
+  | 'rail.check.simulated';
 
 /** Structured event record. */
 export interface MACPEvent {
@@ -88,7 +102,14 @@ export interface GateResult {
   type: string;
   output: string;
   timed_out: boolean;
+  /** Back-compat boolean view — true ONLY when `status === 'passed'`. */
   passed: boolean;
+  /** Typed discriminator — the authoritative gate outcome (RI-N2). */
+  status: GateStatus;
+  /** Typed capability error code, set when `status === 'capability_failure'`. */
+  capability_code?: MacpErrorCode;
+  /** Why a non-executed state (simulated/waiting/capability_failure) was reached. */
+  reason?: string;
   fail_on?: string;
   blockers?: number;
   findings?: number;
@@ -96,6 +117,22 @@ export interface GateResult {
   parse_error?: string;
 }
 
+/**
+ * Aggregate outcome of `runGates` (RI-N2).
+ *
+ * `state` is the typed aggregate: it is `passed` only when every gate really
+ * executed green. A `simulated` result makes the aggregate `simulated` (never
+ * `passed`); a `waiting` manual gate keeps the aggregate `waiting`; a missing
+ * capability makes it `capability_failure`. `allPassed` is exactly
+ * `state === 'passed'`, so a simulated or waiting result can never satisfy a
+ * dependency, acceptance criterion, gate, merge, or release check.
+ */
+export interface RunGatesResult {
+  allPassed: boolean;
+  gateResults: GateResult[];
+  state: GateStatus;
+}
+
 /** Result from a completed task. */
 export interface TaskResult {
   task_id: string;

packages/mosaic/framework/defaults/AGENTS.md

@@ -43,6 +43,8 @@ overwritten on upgrade. (Layer model: `constitution/LAYER-MODEL.md`.)
 | Secrets / vault usage                              | `guides/VAULT-SECRETS.md`          |
 | Tool/credential reference (service CLIs, wrappers) | `guides/TOOLS-REFERENCE.md`        |
 | Memory protocol (OpenBrain capture/recall)         | `guides/MEMORY.md`                 |
+| Seat identity, git credentials, token slots        | `guides/SEAT-IDENTITY.md`          |
+| Reaching another agent (fleet comms)               | `guides/FLEET-COMMS.md`            |
 
 ## Subagent Model Selection (Cost — Hard Rule)
 

packages/mosaic/framework/guides/CODE-REVIEW.md

@@ -15,8 +15,78 @@ Merge strategy enforcement (HARD RULE):
 - Merge to `main` MUST be squash-only.
 - Use `~/.config/mosaic/tools/git/pr-merge.sh -n {PR_NUMBER} -m squash --expect-head {approved_full_sha}` (or PowerShell equivalent).
 
+An estate MAY carry a documented exception for a repository whose gates are commit hooks rather
+than review. Such an exception belongs in that estate's own working copy of this guide, is
+scoped to the named repository, and is never precedent for a second one.
+
+**Do not use `pr-review.sh` or `issue-comment.sh` to post a verdict** (mosaicstack#1280). Post
+through a direct authenticated API call as your own seat, or hand the verdict to the requesting
+seat. Handing it over is a legitimate delivery path, not a fallback.
+
+## Evidence Discipline (applies to every finding)
+
+The checklist below says what to look at. This section says when you are allowed to believe what
+you saw. Every rule here was earned by a wrong conclusion that reached a report.
+
+1. **A finding is a claim about behavior.** State the failing input, the path taken, and the
+   wrong result. "This looks fragile" is not a finding.
+2. **A green check is not a result until you have shown it could go red.** Run the control. A
+   `0`, an empty result, or a column of identical values with no failing counterpart is a
+   non-result.
+3. **Measurement and explanation are separate sentences.** Report the command and its output,
+   then, as its own sentence, what you think it means.
+4. **Never widen the case you measured.** If you checked one path, the finding covers one path.
+5. **Reproduce a reported failure before recording it, and say which tree you measured.** Two
+   correct measurements of two different trees disagree without either being wrong.
+6. **Verify by content on the ref that ships**, never by ancestry of a local sha. A rebase mints
+   new shas; a commit being an ancestor of something local proves nothing about the remote.
+   Compare by digest against `origin/<branch>`.
+7. **Confidence is part of the finding.** "I could not reproduce this" is a usable review
+   comment. A confident guess is not.
+8. **Author is not reviewer** (Gate-16). Do not review your own work, or work you shaped closely
+   enough to be a co-author of. Say so and hand it back.
+
+### Measuring a shell suite
+
+Each of these produced a wrong conclusion before it was written down.
+
+9. **`cmd | tail; echo rc=$?` reports `tail`'s exit code, not `cmd`'s.** It reads as a pass when
+   the command failed. Redirect to a file and check `rc` directly, or use `${PIPESTATUS[0]}`.
+10. **Under `set -o pipefail`, a missed glob makes `ls` exit 2**, the pipeline inherits it, and
+    `set -e` kills the run. Iterate a glob with a `for` loop and an `-e` test instead of piping
+    `ls`.
+11. **A suite that exits nonzero with ZERO output is an environment question, not a defect in
+    the code under review.** The usual cause is a sourced dependency that is absent, so `set -e`
+    kills the first case before anything prints. Extract whole tool trees — `tools/git` alone is
+    missing `tools/_lib/credentials.sh`. Isolate the variable and prove it by adding only that
+    back.
+12. **`git -C <dir>` in a directory that is not itself a repo answers from the enclosing repo.**
+    A scratch tree under `~/.mosaic` reports `~/.mosaic`'s HEAD, not the PR's, and every
+    conclusion drawn from it describes the wrong tree. Confirm `git rev-parse --show-toplevel`
+    is the tree you think it is before trusting any git output.
+
+### Feedback Categories
+
+- **Blocker**: must fix before merge (security, bugs, test failures)
+- **Should Fix**: important but not blocking (code quality, minor issues)
+- **Suggestion**: optional improvement (style preference, nice-to-have)
+- **Question**: seeking clarification
+
 ## Review Checklist
 
+Reviewer seats split this checklist by class rather than duplicating it. A seat reviews its own
+sections in full and may raise anything it notices outside them as a Suggestion, never as a
+Blocker on someone else's ground.
+
+| Reviewer class   | Owns                                                                                                    |
+| ---------------- | ------------------------------------------------------------------------------------------------------- |
+| `rev-code-*`     | 1 Correctness, 3 Testing, 4 Code Quality, 4a TypeScript, 5 Documentation, 6 Performance, 7 Dependencies |
+| `rev-security-*` | 2 Security, 2a OWASP                                                                                    |
+
+Where two seats of the same class review the same change, they review independently and compare
+after. A second seat that reads the first seat's findings before measuring is a proofreader, not
+a second opinion.
+
 ### 1. Correctness
 
 - [ ] Code does what the issue/PR description says
@@ -53,7 +123,7 @@ Merge strategy enforcement (HARD RULE):
 - [ ] Tests cover happy path AND error cases
 - [ ] Situational tests cover all impacted change surfaces (primary gate)
 - [ ] Tests validate required behavior/outcomes, not only internal implementation details
-- [ ] TDD was applied when required by `~/.config/mosaic/guides/QA-TESTING.md`
+- [ ] TDD was applied when required by `guides/QA-TESTING.md`
 - [ ] Coverage meets 85% minimum
 - [ ] Tests are readable and maintainable
 - [ ] No flaky tests introduced
@@ -82,7 +152,7 @@ Merge strategy enforcement (HARD RULE):
 ### 5. Documentation
 
 - [ ] Complex logic has explanatory comments
-- [ ] Required docs updated per `~/.config/mosaic/guides/DOCUMENTATION.md`
+- [ ] Required docs updated per `guides/DOCUMENTATION.md`
 - [ ] Public APIs are documented
 - [ ] Private/internal APIs are documented
 - [ ] API input/output schemas are documented
@@ -126,13 +196,6 @@ git diff main...HEAD
 - Distinguish between blocking issues and suggestions
 - Be constructive, not critical of the person
 
-### Feedback Categories
-
-- **Blocker**: Must fix before merge (security, bugs, test failures)
-- **Should Fix**: Important but not blocking (code quality, minor issues)
-- **Suggestion**: Optional improvements (style preferences, nice-to-haves)
-- **Question**: Seeking clarification
-
 ### Review Comment Format
 
 ```

packages/mosaic/framework/guides/FLEET-COMMS.md

@@ -0,0 +1,86 @@
+# Fleet Comms Guide
+
+How one seat reaches another on a host. The mechanism is the framework's; the sessions and
+sockets are per-host, so measure yours rather than trusting an example.
+
+`mosaic <runtime>` would normally inject the addressing block from the roster. Where the composer
+is unavailable, or where the roster is stale, this guide is the substitute.
+
+## Measure the fleet; do not trust the roster
+
+`fleet/roster.yaml` is a declaration of intent, not an observation. It routinely names a socket
+that was never created, lists seats that are not running, and omits seats that are — this was
+all three have been observed true at once on a live host. Find out what is actually
+up before addressing anyone:
+
+```bash
+tmux list-sessions
+tmux list-panes -a -F '#{session_name}  #{pane_current_command}  #{pane_current_path}'
+```
+
+The pane command tells you the runtime. A pane showing `bash` is an idle shell with no agent
+attached — a send there lands in a shell prompt and is not read by anyone.
+
+Use the **default socket**. Do not pass `-L mosaic-fleet` on the strength of the roster.
+
+## Sending
+
+```bash
+~/.config/mosaic/tools/tmux/agent-send.sh -s <dst_session> -C <class> -m "<message>"
+```
+
+`-s` also accepts `session:window.pane`. `-f <file>` sends a file body; stdin works too.
+
+### Classes
+
+`-C` takes exactly one of these. Anything else exits 3.
+
+| Class          | Use for                                                  |
+| -------------- | -------------------------------------------------------- |
+| `terminal-log` | log only; never needs the agent's attention              |
+| `actionable`   | a decision, blocker, gate, or question needing an answer |
+| `human`        | relayed from a human operator                            |
+| `reaction`     | an ack or acknowledgement token                          |
+| `digest`       | machine wake, coalescible                                |
+
+An absent class is treated as `actionable` by consumers, which is the fail-safe direction. Prefer
+naming it anyway.
+
+### Addressing preamble
+
+The wire format is `[<src> -> <dst> class=<class>] <body>`. Flip it when you reply — the tool
+sends, it does not auto-reply.
+
+### Exit codes
+
+| rc  | Meaning                                        |
+| --- | ---------------------------------------------- |
+| 0   | delivered or queued                            |
+| 1   | target session not found                       |
+| 2   | text reached the pane but is **still a draft** |
+| 3   | usage error (bad class, missing `-s`)          |
+
+**Never retry on rc=2.** The message is in the target pane; retrying double-sends it. Confirm
+instead:
+
+```bash
+tmux capture-pane -p -t <session>:0.0 | tail -20
+```
+
+rc=2 is the normal result when the target is an idle pi seat.
+
+## Durable comms
+
+tmux delivery is host-local and does not survive a pane. Anything that must outlive the session
+goes through the estate's durable comms protocol — a committed `comms/` tree in an estate repo,
+with its own README. Use it for cross-host messages, verdicts, and anything a later session needs
+to find.
+
+## Handing work across seats
+
+1. **A verdict handed to the requesting seat is a legitimate delivery path**, and the required one
+   for anything `pr-review.sh` would otherwise post (see `guides/CODE-REVIEW.md`).
+2. **Address the seat, not the runtime.** A seat name is a session name; whether it runs claude,
+   pi or codex is not the sender's business.
+3. **Say what you measured, not just what you concluded** — the receiving seat cannot see your
+   terminal.

packages/mosaic/framework/guides/SEAT-IDENTITY.md

@@ -0,0 +1,133 @@
+# Seat Identity & Credentials Guide
+
+Every agent that touches a Mosaic-managed git host acts as a named seat with its own credential.
+This guide is how that works on a host, and what an agent must never do with it.
+
+The mechanism below is the framework's. The specific paths, seats and stores are per-host:
+measure yours before trusting any of them.
+
+## The rule
+
+**One seat, one identity, one token file.** A seat never borrows another seat's credential, never
+falls back to a shared owner account, and never carries a second copy of its own token. A second
+copy is drift, and drift surfaces as the stale copy returning 401 — which reads as a revoked
+token and sends whoever debugs it somewhere else entirely.
+
+A credential refusal is correct behavior, not a bug to route around. If git refuses with a
+fail-closed diagnostic, the fix is to provision or correct _your_ identity. Escalate; do not
+substitute.
+
+## How a credential is resolved
+
+Find the helper the way **git** does, not with `command -v`. Git runs whatever
+`credential.helper` names, and on a Mosaic host that is an absolute path — so a PATH lookup
+answers a different question and the two disagree the moment the PATH copy is removed. It was
+removed on hosts that have completed that migration.
+
+```bash
+git config --get-all credential.helper     # every helper, in the order git tries them
+```
+
+Git tries **each** configured helper in turn until one supplies a credential. A fail-closed
+helper supplies nothing, so a second helper configured behind it silently becomes the one that
+answers. When you care which binary serves a credential, read the whole list.
+Resolve all three forms git accepts — absolute path, `!command`, and a bare name looked up on
+PATH — not just the one your host happens to use.
+
+The helper resolves the identity in this order:
+
+1. `$MOSAIC_GIT_IDENTITY`
+2. `git config --get mosaic.gitIdentity`
+3. the username git supplied on stdin
+
+It maps the host to a store prefix — `git.mosaicstack.dev` to `gitea-mosaicstack`,
+`git.uscllc.com` to `gitea-usc`. Any other host is declined quietly with rc=0, which is not an
+error and raises no escalation.
+
+Then it chooses **one** of two stores, and reads exactly one file:
+
+```
+brain_home = ${MOSAIC_BRAIN_HOME:-$HOME/.mosaic}
+
+seat     — when $brain_home/fleet/agents/<identity>/ EXISTS
+           $brain_home/fleet/agents/<identity>/secrets/<prefix>-<identity>.token
+service  — otherwise
+           ~/.config/mosaic/secrets/gitea-tokens/<prefix>-<identity>.token
+```
+
+**There is no precedence between the two and no fallback from one to the other.** The existence
+of the seat directory decides it. A seat that has a directory and an empty slot fails closed; it
+does not reach the service store. That is the intended behavior — the alternative is an agent
+silently acting as somebody else.
+
+If the file is unreadable the helper **fails closed**: it refuses and writes a durable record to
+the escalation spool. It does not fall back to a shared account. The record is what exists — any
+alerting built on top of it is a separate, best-effort concern and is not performed by the helper,
+so do not wait for a notification that nothing sends. That fallback is what made
+`usc/uconnect#3084` unattributable, and it was removed deliberately.
+
+Verify the helper you actually have:
+
+```bash
+h=$(git config --get credential.helper)
+grep -c 'FAIL CLOSED'  "$h"    # expect >= 1
+grep -c 'fleet/agents' "$h"    # expect >= 1; 0 means it predates mosaicstack#1311
+```
+
+## Where a seat's token lives
+
+The seat slot is the **only** copy:
+
+```
+~/.mosaic/fleet/agents/<seat>/secrets/<prefix>-<seat>.token      real file, mode 600
+```
+
+The framework store at `~/.config/mosaic/secrets/gitea-tokens/` holds tokens for **service
+identities only** — identities with no seat directory. A seat's token does not belong there.
+
+Before mosaicstack#1311 the deployed helper knew only the service store, and seats were bridged
+with a symlink from the store into the slot. **Those bridges must be removed once a seat-aware helper is deployed, and must not be
+recreated.** Remove them only after the helper can reach the slot without them; the reverse order
+takes every seat offline. A symlink
+is not how a system finds a credential; the helper resolving the right store is.
+
+`.principal` and `.scopes` beside the token are grant records, not secrets. They are tracked. The
+`.token` never is.
+
+### Provisioning a new seat
+
+1. Create `~/.mosaic/fleet/agents/<seat>/secrets/` mode 700.
+2. Write `.principal` (the Gitea login) and `.scopes` (the granted scopes), mode 600.
+3. The estate operator mints the token into the seat slot, mode 600. Agents do not mint their
+   own, and do not ask another agent to mint one for them.
+4. Verify with an authenticated `GET /user` and confirm the returned login is the seat, **not the
+   minting account**. Record the date in `ENTITY.md`. Never record the value.
+
+There is no step that links the framework store to the slot. A seat-aware helper reads the slot
+directly; a store entry pointing at a slot is the bridge described in **Where a seat's token lives** above,
+and it is not part of provisioning.
+
+Until step 3, the seat is unminted and its git writes fail closed. That is the designed state and
+is safe to launch in — the seat is told at launch so it does not discover it mid-task.
+
+## Acting as yourself
+
+Name the identity on every invocation:
+
+```bash
+MOSAIC_GIT_IDENTITY=<seat> git push
+git -c user.name=<seat> -c user.email=<seat>@mosaicstack.dev commit -m "..."
+```
+
+**Never persist `git config mosaic.gitIdentity` inside a `~/src/stack` worktree.** Every worktree
+of that clone shares one `.git/config`, so a persisted identity there silently rewrites the
+identity of every other seat working in that clone. The per-invocation form has no exception.
+
+## Handling
+
+1. **Never print a token value.** Compare by SHA-256 digest, or write `<REDACTED>`.
+2. **Never stage a `.token`, `secrets.json`, or `ENTITY.md`.** Stage explicit paths and **never
+   `git add -A`** — `secrets/*.principal` and `secrets/*.scopes` are covered by no ignore rule.
+3. **Never place a token in an environment variable** in an interactive session. A `declare -x`
+   dump has leaked the whole environment to a terminal before.
+4. **No real credential or operator data on a sandbox VM, ever.**

packages/mosaic/framework/templates/agent/fragments/conditional-loading.md

@@ -15,3 +15,5 @@
 | Infrastructure/DevOps                    | `~/.config/mosaic/guides/INFRASTRUCTURE.md` |
 | QA/Testing                               | `~/.config/mosaic/guides/QA-TESTING.md`     |
 | Secrets management (Vault)               | `~/.config/mosaic/guides/VAULT-SECRETS.md`  |
+| Seat identity / git credentials          | `~/.config/mosaic/guides/SEAT-IDENTITY.md`  |
+| Reaching another agent (fleet comms)     | `~/.config/mosaic/guides/FLEET-COMMS.md`    |

packages/mosaic/framework/tools/_scripts/mosaic-doctor

@@ -51,6 +51,8 @@ fix() { fix_count=$((fix_count + 1)); echo "[FIX] $*"; }
 
 warn_count=0
 warn() { warn_count=$((warn_count + 1)); echo "[WARN] $*"; }
+note() { echo "[NOTE] $*"; return 0; }
+
 pass() {
   if [[ $VERBOSE -eq 1 ]]; then
     echo "[OK] $*"
@@ -288,6 +290,18 @@ check_brain_home() {
   brain="$(resolve_brain_home)"
 
   if [[ "$brain" == "$MOSAIC_HOME" ]]; then
+    # Implicit-path greenfield case (#1288 comment 23133, fred's trace): nothing
+    # in product code creates ~/.mosaic/fleet/agents — the first fleet write
+    # resolves legacy (generated-env-boundary resolves before creating) and
+    # then manufactures the evidence that keeps the host legacy. On a host with
+    # ~/.mosaic but no fleet/agents, the three operator checks all agree and all
+    # point the wrong way; this doctor is the only one that can disagree, so it
+    # must say it — as a note, not a warn: nothing is broken yet.
+    if [[ "$(cd "$MOSAIC_HOME" 2>/dev/null && pwd -P)" == "$HOME/.config/mosaic" \
+       && -d "$HOME/.mosaic" && ! -d "$HOME/.mosaic/fleet/agents" ]]; then
+      note "Fleet state home: $MOSAIC_HOME (legacy). NOTE: ~/.mosaic exists but carries no fleet/agents — the first 'mosaic fleet regen' on this host locks in the legacy tree. Create ~/.mosaic/fleet/agents first to adopt the brain."
+      return
+    fi
     pass "Fleet state home: $MOSAIC_HOME (legacy single-tree; no brain adopted)"
     return
   fi

packages/mosaic/framework/tools/_scripts/test-brain-home-check.sh

@@ -39,22 +39,26 @@ eval "$(extract_function check_brain_home)"
 ROOT=$(mktemp -d)
 trap 'rm -rf "$ROOT"' EXIT
 
+# note output is neither [OK] nor [WARN] — assert it directly in the case below.
 run_case() {
-  # label, expect (ok|warn), then env assignments as arguments.
+  # label, expect (ok|warn|note), then env assignments as arguments.
   # The check runs under `env` in a subshell, so its warn() also prints a
   # sentinel the parent counts — a subshell counter would never be visible.
   local label="$1" expect="$2"
   shift 2
-  local out warns
+  local out warns notes
-  out=$(env "$@" bash -c "warn() { echo \"[WARN] \$*\"; }; pass() { echo \"[OK] \$*\"; return 0; }; $(extract_function resolve_brain_home); $(extract_function check_brain_home); check_brain_home" 2>&1)
+  out=$(env "$@" bash -c "warn() { echo \"[WARN] \$*\"; }; note() { echo \"[NOTE] \$*\"; return 0; }; pass() { echo \"[OK] \$*\"; return 0; }; $(extract_function resolve_brain_home); $(extract_function check_brain_home); check_brain_home" 2>&1)
   warns=$(printf '%s\n' "$out" | grep -c '^\[WARN\]' || true)
-  if [[ "$expect" == ok && "$warns" -eq 0 ]]; then
+  notes=$(printf '%s\n' "$out" | grep -c '^\[NOTE\]' || true)
+  if [[ "$expect" == ok && "$warns" -eq 0 && "$notes" -eq 0 ]]; then
     echo "ok - $label"
   elif [[ "$expect" == warn && "$warns" -gt 0 ]]; then
     echo "ok - $label (warned)"
+  elif [[ "$expect" == note && "$notes" -gt 0 ]]; then
+    echo "ok - $label (noted)"
   else
     echo "output: $out" >&2
-    fail "$label: expected $expect (warns=$warns)"
+    fail "$label: expected $expect (warns=$warns notes=$notes)"
   fi
 }
 
@@ -105,4 +109,19 @@ chmod 700 "$ROOT/brain-clean/fleet/agents" "$ROOT/config6/fleet/agents"
 run_case "empty config-home agents dir alongside brain passes" ok \
   MOSAIC_HOME="$ROOT/config6" HOME="$ROOT" MOSAIC_BRAIN_HOME="$ROOT/brain-clean"
 
+# ── greenfield brain-without-agents at the default home → note (#1288) ─────
+mkdir -p "$ROOT/gf-home/.config/mosaic/fleet" "$ROOT/gf-home/.mosaic"
+run_case "~/.mosaic without fleet/agents at default home notes the lock-in" note \
+  MOSAIC_HOME="$ROOT/gf-home/.config/mosaic" HOME="$ROOT/gf-home"
+
+# ── no ~/.mosaic at all at the default home → clean pass ─────────────────
+mkdir -p "$ROOT/plain-home/.config/mosaic/fleet"
+run_case "no ~/.mosaic at default home passes silently" ok \
+  MOSAIC_HOME="$ROOT/plain-home/.config/mosaic" HOME="$ROOT/plain-home"
+
+# ── custom (non-default) home with a stray ~/.mosaic → still silent ──────
+mkdir -p "$ROOT/custom-home/fleet/agents" "$ROOT/custom-home/.mosaic"
+run_case "custom home with stray ~/.mosaic stays silent" ok \
+  MOSAIC_HOME="$ROOT/custom-home" HOME="$ROOT/custom-home"
+
 echo "ok - mosaic-doctor brain-home check"

packages/mosaic/framework/tools/git/README.md

@@ -43,7 +43,7 @@ fleet commits, pushes, and opens PRs under one identity — with no cryptographi
 separation between an author and a reviewer.
 
 Both `git-credential-mosaic` and `get_gitea_token()` resolve an optional **per-agent
-identity** before falling back to the shared account:
+identity**:
 
 1. `MOSAIC_GIT_IDENTITY` environment variable, or
 2. `git config --get mosaic.gitIdentity` (set per-worktree; persists on disk across
@@ -51,12 +51,54 @@ identity** before falling back to the shared account:
 3. (git-credential-mosaic only) the username git itself supplies for the credential
    request.
 
-If the resolved identity has a token file at
+### Which store a credential is read from
-`~/.config/mosaic/secrets/gitea-tokens/gitea-{usc,mosaicstack}-<agent-id>.token`, that
+
-identity + token is used. **Nothing configured → nothing changes**: with no per-slot
+The store is chosen by what the identity **is**, not by which file happens to exist first:
-token file present, both tools fall through to the existing shared-account path
+
-unchanged, so this feature is a no-op on any host that hasn't provisioned per-slot
+| The identity                                                       | Its credential is read from                                                |
-tokens.
+| ------------------------------------------------------------------ | -------------------------------------------------------------------------- |
+| has a directory at `<brain>/fleet/agents/<id>/` — it is a **seat** | `<brain>/fleet/agents/<id>/secrets/gitea-{usc,mosaicstack}-<id>.token`     |
+| does not — it is a **service identity**                            | `~/.config/mosaic/secrets/gitea-tokens/gitea-{usc,mosaicstack}-<id>.token` |
+
+`<brain>` is `MOSAIC_BRAIN_HOME` if set, else `~/.mosaic` — the same resolution
+`packages/mosaic/src/fleet/brain-home.ts` performs.
+
+**There is no precedence between the two stores and no fallback from one to the other.**
+A seat whose slot is empty is refused even when a same-named token sits in the framework
+store. One credential lives in exactly one location: a second copy is drift rather than
+redundancy, and the way drift surfaces is a stale copy returning 401, which reads as a
+revoked token and sends whoever debugs it to the wrong place.
+
+### What happens when nothing resolves
+
+| identity resolves | token in its store | host runs a fleet | result                    |
+| ----------------- | ------------------ | ----------------- | ------------------------- |
+| yes               | yes                | —                 | that identity + token     |
+| yes               | no                 | —                 | **fail closed**           |
+| no                | —                  | yes               | **fail closed**           |
+| no                | —                  | no                | shared account, unchanged |
+
+A host "runs a fleet" when `<brain>/fleet/agents` exists — the same signal `brain-home.ts`
+uses to decide a brain is active.
+
+Failing closed means: nothing is emitted, the exit status is nonzero, a stderr diagnostic
+names the identity, its source, the store it resolved to and the path that was expected,
+and `git-credential-mosaic` additionally appends a record (identity, host, reason, cwd —
+never a token value) to `${MOSAIC_CREDENTIAL_SPOOL:-~/.local/state/mosaic-credential-escalations}`.
+The git operation fails; nothing is attributed to anyone.
+
+The shared-account fallback that used to cover these two cases is why a PR could be
+authored, commented and merged under an account whose owner did not open it — every seat
+shared one identity, so the record could not be traced back afterwards. An
+under-provisioned agent is refused rather than handed the most privileged account
+available.
+
+**On a host with no fleet, nothing changes**: no `fleet/agents` directory means the shared
+account still answers, so this is a no-op for an operator who has not provisioned per-slot
+tokens. On a host that does run a fleet, a human doing manual git work needs an identity
+of their own — `MOSAIC_GIT_IDENTITY=<id>` with a provisioned slot. There is deliberately no
+environment variable that restores the fallback; one would reintroduce exactly the
+substitution this removes.
 
 ### Enabling it for a clone
 

packages/mosaic/framework/tools/git/detect-platform.sh

@@ -507,37 +507,69 @@ get_gitea_token() {
 
     # 0. Per-agent identity (Gate-16 author≠reviewer). If MOSAIC_GIT_IDENTITY, or the
     #    per-worktree `git config mosaic.gitIdentity`, resolves to an agent that has a
-    #    stored per-slot token for this host, act AS that agent so API tooling
+    #    stored credential for this host, act AS that agent so API tooling
     #    (pr-create, issue-create, …) authors under the right identity — matching the
-    #    git credential helper. Backward-compatible: nothing resolvable → shared logic below.
+    #    git credential helper, which this block deliberately mirrors.
     local _ident="${MOSAIC_GIT_IDENTITY:-}"
     local _ident_src="MOSAIC_GIT_IDENTITY"
     if [[ -z "$_ident" ]]; then
         _ident="$(git config --get mosaic.gitIdentity 2>/dev/null || true)"
         _ident_src="git config mosaic.gitIdentity"
     fi
-    if [[ -n "$_ident" ]]; then
+
-        local _idpfx=""
+    # Recognized Gitea hosts are the ones carrying the per-identity token scheme.
-        case "$host" in
+    local _idpfx=""
-            git.uscllc.com)      _idpfx=gitea-usc ;;
+    case "$host" in
-            git.mosaicstack.dev) _idpfx=gitea-mosaicstack ;;
+        git.uscllc.com)      _idpfx=gitea-usc ;;
-        esac
+        git.mosaicstack.dev) _idpfx=gitea-mosaicstack ;;
-        if [[ -n "$_idpfx" ]]; then
+    esac
-            local _idtok="$HOME/.config/mosaic/secrets/gitea-tokens/${_idpfx}-${_ident}.token"
+
-            if [[ -r "$_idtok" ]]; then
+    # Brain-home resolution mirrors packages/mosaic/src/fleet/brain-home.ts and
-                cat "$_idtok"
+    # tools/fleet/start-agent-session.sh: MOSAIC_BRAIN_HOME wins, else ~/.mosaic.
-                return 0
+    local _brain_home="${MOSAIC_BRAIN_HOME:-$HOME/.mosaic}"
-            fi
+
-            # FAIL LOUD: an explicit git identity was requested for a recognized Gitea host,
+    if [[ -n "$_ident" && -n "$_idpfx" ]]; then
-            # but no per-slot token exists for THAT identity. Refuse to fall through to the
+        # Credential store selection: an identity is a SEAT or a SERVICE, and which
-            # shared/default credential loader below — silently borrowing another slot's token
+        # one it is decides where its credential lives. No precedence between the
-            # would post PRs/issues/reviews under the WRONG agent (e.g. rev2's review attributed
+        # two stores and no fallback from one to the other — a seat with an empty
-            # to coder3), corrupting Gate-16 author≠reviewer separation. Hard-stop instead so the
+        # slot fails loud rather than reading a service credential of the same name.
-            # caller aborts loudly rather than acting as the wrong identity.
+        # One credential, one location: two copies diverge, and the stale copy fails
-            echo "Error: git identity '$_ident' requested (via $_ident_src) for host '$host', but no per-slot token at $_idtok." >&2
+        # in a way that reads as a revoked token rather than as drift.
-            echo "       Refusing to borrow another slot's token. Provision the per-slot token, or unset the identity to use shared credentials." >&2
+        local _idtok _ident_kind
-            return 1
+        if [[ -d "$_brain_home/fleet/agents/$_ident" ]]; then
+            _ident_kind="seat"
+            _idtok="$_brain_home/fleet/agents/$_ident/secrets/${_idpfx}-${_ident}.token"
+        else
+            _ident_kind="service identity"
+            _idtok="$HOME/.config/mosaic/secrets/gitea-tokens/${_idpfx}-${_ident}.token"
         fi
+        if [[ -r "$_idtok" ]]; then
+            cat "$_idtok"
+            return 0
+        fi
+        # FAIL LOUD: an explicit git identity was requested for a recognized Gitea host,
+        # but no credential exists for THAT identity. Refuse to fall through to the
+        # shared/default credential loader below — silently borrowing another identity's
+        # token would post PRs/issues/reviews under the WRONG agent (e.g. rev2's review
+        # attributed to coder3), corrupting Gate-16 author≠reviewer separation. Hard-stop
+        # instead so the caller aborts loudly rather than acting as the wrong identity.
+        echo "Error: git identity '$_ident' requested (via $_ident_src) for host '$host', but no per-slot token at $_idtok." >&2
+        echo "       Resolved as a ${_ident_kind}; there is no fallback between the seat and service stores." >&2
+        echo "       Refusing to borrow another slot's token. Provision the credential at that path, or unset the identity to use shared credentials." >&2
+        return 1
+    fi
+
+    # FAIL LOUD: no identity resolved, on a host that HAS a fleet. Where seats exist,
+    # every record must name the agent that made it, so an unattributable request is
+    # refused rather than handed the shared account. `fleet/agents` existing is the
+    # same signal brain-home.ts uses to decide a brain is active. A host with no fleet
+    # keeps the shared path below unchanged: there the shared account is the operator's
+    # own and there is no attribution to lose.
+    if [[ -z "$_ident" && -n "$_idpfx" && -d "$_brain_home/fleet/agents" ]]; then
+        echo "Error: no git identity resolved for host '$host', but this host runs a fleet ($_brain_home/fleet/agents)." >&2
+        echo "       Refusing to fall back to the shared account: records it creates cannot be attributed to the agent that made them." >&2
+        echo "       Set MOSAIC_GIT_IDENTITY=<agent-id> or 'git config mosaic.gitIdentity <agent-id>'." >&2
+        return 1
     fi
 
     # 1. Mosaic credential loader (host → service mapping, run in subshell to avoid polluting env)

packages/mosaic/framework/tools/git/git-credential-mosaic

@@ -1,21 +1,48 @@
 #!/bin/bash
-# git-credential-mosaic — git credential helper — resolves Gitea tokens from
+# git-credential-mosaic — git credential helper. Resolves a Gitea token from the
-# the Mosaic credential store at runtime so remote URLs never embed secrets.
+# Mosaic credential store at runtime so remote URLs never embed secrets.
 #
 # Install (one-time, per clone or globally):
 #   git config credential.helper "$HOME/.config/mosaic/tools/git/git-credential-mosaic"
-#   # or, fleet-wide: git config --global credential.helper "$HOME/.config/mosaic/tools/git/git-credential-mosaic"
 #
-# Per-agent Gate-16 identity (author != reviewer separation):
+# Per-agent identity (Gate-16 author != reviewer separation):
 #   git config mosaic.gitIdentity <agent-id>   # per-worktree, persists on disk
 #   # or: export MOSAIC_GIT_IDENTITY=<agent-id>
 #
-# Resolution priority: MOSAIC_GIT_IDENTITY env > git config mosaic.gitIdentity
+# ── WHY THIS FAILS CLOSED ──────────────────────────────────────────────────────
-# (per-worktree, survives across non-persistent shells) > git-supplied username
+# This helper used to end by emitting the shared account's token for any request
-# (credential.username / URL). When the resolved identity has a matching
+# it could not resolve to an identity. A seat with no identity, or with an
-# per-agent token file, use it instead of the shared account. Backward
+# identity whose token was never provisioned, therefore received the most
-# compatible: nothing resolvable -> shared token (unchanged behavior).
+# privileged credential configured on the host — silently, and indistinguishably
+# from correct operation. Every record it then created (commit, push, PR, review)
+# was attributed to that shared account, so author != reviewer separation was
+# unenforceable and the true actor was unrecoverable after the fact.
+#
+# Under-provisioning must fail loudly, not impersonate. A refused git operation
+# is recoverable in one command; a merged pull request attributed to the wrong
+# principal is not.
+#
+# ── CONTRACT ───────────────────────────────────────────────────────────────────
+#   identity : MOSAIC_GIT_IDENTITY  >  git config mosaic.gitIdentity  >  the
+#              username git supplies on stdin
+#   store    : chosen by what the identity IS, with no precedence and no
+#              cross-store fallback (see "Credential store selection" below)
+#   hit      : emit username + password, exit 0
+#   miss     : emit NOTHING, spool a durable escalation record, explain on
+#              stderr, exit 1 — git surfaces the failure and nothing is attributed
+#   unknown host : exit 0 with no output, no record (passthrough for non-Mosaic
+#              remotes handled by another helper)
+#
+# Backward compatibility is preserved for exactly one case: a host with no fleet
+# and no identity requested still gets the shared account, because on such a host
+# the shared account is the operator's own and there is no attribution to lose.
+# A host that HAS a fleet has agents whose records must be distinguishable, so
+# the shared fallback is refused there.
+#
+# A token is never written to stderr, to the escalation record, or to any log.
+
 [ "$1" = "get" ] || exit 0
+
 host=""; username_in=""
 while IFS= read -r line; do
   [ -z "$line" ] && break
@@ -24,46 +51,170 @@ while IFS= read -r line; do
     username=*) username_in=${line#username=};;
   esac
 done
-# Per-agent identity resolution (Gate-16 author≠reviewer separation).
+
-# Priority: MOSAIC_GIT_IDENTITY env  >  git config mosaic.gitIdentity (per-worktree,
+# Recognized Gitea hosts carry the per-identity token scheme. Anything else is
-# survives across non-persistent shells)  >  git-supplied username (credential.username
+# declined quietly — another helper owns it, and refusing would break it.
-# / URL). When the resolved identity has a matching per-agent token, use it instead of
-# the shared account. Backward-compatible: nothing resolvable → shared token.
-ident="$MOSAIC_GIT_IDENTITY"
-[ -z "$ident" ] && ident=$(git config --get mosaic.gitIdentity 2>/dev/null)
-[ -z "$ident" ] && ident="$username_in"
-if [ -n "$ident" ]; then
-  case "$host" in
-    git.uscllc.com)      idpfx=gitea-usc;;
-    git.mosaicstack.dev) idpfx=gitea-mosaicstack;;
-    *) idpfx="";;
-  esac
-  if [ -n "$idpfx" ]; then
-    idtok="$HOME/.config/mosaic/secrets/gitea-tokens/${idpfx}-${ident}.token"
-    if [ -r "$idtok" ]; then
-      echo "username=${ident}"
-      echo "password=$(cat "$idtok")"
-      exit 0
-    fi
-  fi
-fi
 case "$host" in
-  git.uscllc.com)      svc=gitea-usc;;
+  git.uscllc.com)      idpfx=gitea-usc;;
-  git.mosaicstack.dev) svc=gitea-mosaicstack;;
+  git.mosaicstack.dev) idpfx=gitea-mosaicstack;;
   *) exit 0;;
 esac
-# Script-relative (not $HOME-absolute) so this resolves correctly regardless
+
-# of where the framework installer places tools/ under $HOME — mirrors
+ident="$MOSAIC_GIT_IDENTITY"; ident_src="MOSAIC_GIT_IDENTITY"
-# detect-platform.sh's own cred_loader resolution in this same directory.
+if [ -z "$ident" ]; then
-script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+  ident=$(git config --get mosaic.gitIdentity 2>/dev/null)
-# shellcheck source=../_lib/credentials.sh
+  ident_src="git config mosaic.gitIdentity"
-source "$script_dir/../_lib/credentials.sh"
+fi
-load_credentials "$svc" >/dev/null 2>&1 || exit 0
+if [ -z "$ident" ]; then
-# GITEA_USER is not populated by load_credentials (it only exports
+  ident="$username_in"
-# GITEA_URL/GITEA_TOKEN for gitea-*), so this fallback is normally taken. Gitea's
+  ident_src="the username git supplied"
-# git-over-HTTP auth authenticates from the token itself (the password field),
+fi
-# not from the username string, so any non-empty placeholder works here — this
+
-# is deliberately NOT a real account name (framework files must stay
+# ── Credential store selection ────────────────────────────────────────────────
-# operator-agnostic; see tools/quality/scripts/verify-sanitized.sh).
+# An identity is a SEAT or it is a SERVICE, and which one it is determines where
-echo "username=${GITEA_USER:-git}"
+# its credential lives. There is no precedence rule between the two stores and no
-echo "password=$GITEA_TOKEN"
+# fallback from one to the other: a seat whose slot is empty fails closed rather
+# than reading a service credential that happens to share its name.
+#
+#   seat     — <brain>/fleet/agents/<ident>/ exists
+#              credential at <brain>/fleet/agents/<ident>/secrets/<idpfx>-<ident>.token
+#   service  — it does not
+#              credential at ~/.config/mosaic/secrets/gitea-tokens/<idpfx>-<ident>.token
+#
+# One credential, one location. Two copies of one credential diverge, and the
+# stale copy fails in a way that reads as a revoked token rather than as drift.
+#
+# Brain-home resolution mirrors packages/mosaic/src/fleet/brain-home.ts and
+# tools/fleet/start-agent-session.sh: MOSAIC_BRAIN_HOME wins, else ~/.mosaic.
+brain_home="${MOSAIC_BRAIN_HOME:-$HOME/.mosaic}"
+svc_store="$HOME/.config/mosaic/secrets/gitea-tokens"
+
+idtok=""; ident_kind=""
+if [ -n "$ident" ]; then
+  if [ -d "$brain_home/fleet/agents/$ident" ]; then
+    ident_kind="seat"
+    idtok="$brain_home/fleet/agents/$ident/secrets/${idpfx}-${ident}.token"
+  else
+    ident_kind="service identity"
+    idtok="$svc_store/${idpfx}-${ident}.token"
+  fi
+  if [ -r "$idtok" ]; then
+    echo "username=${ident}"
+    echo "password=$(cat "$idtok")"
+    exit 0
+  fi
+fi
+
+# ── Shared-account fallback: ONLY on a host with no fleet and no identity ──────
+# `fleet/agents` existing is the same signal brain-home.ts uses to decide a brain
+# is active. Where there are seats, records must be attributable, so an
+# unresolvable request is refused instead of borrowing the shared account.
+fleet_present=0
+[ -d "$brain_home/fleet/agents" ] && fleet_present=1
+
+if [ -z "$ident" ] && [ "$fleet_present" -eq 0 ]; then
+  script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+  # shellcheck source=../_lib/credentials.sh
+  source "$script_dir/../_lib/credentials.sh"
+  load_credentials "$idpfx" >/dev/null 2>&1 || exit 0
+  # GITEA_USER is not populated by load_credentials (it exports GITEA_URL and
+  # GITEA_TOKEN only). Gitea's git-over-HTTP auth authenticates from the token in
+  # the password field, not from the username string, so any non-empty
+  # placeholder works — deliberately NOT a real account name, since framework
+  # files stay operator-agnostic (tools/quality/scripts/verify-sanitized.sh).
+  echo "username=${GITEA_USER:-git}"
+  echo "password=$GITEA_TOKEN"
+  exit 0
+fi
+
+# ── FAIL CLOSED ───────────────────────────────────────────────────────────────
+if [ -z "$ident" ]; then
+  reason="no-identity"
+else
+  reason="no-token-for-identity"
+fi
+
+seat="${MOSAIC_AGENT_NAME:-unknown}"
+ts=$(date -u +%Y-%m-%dT%H:%M:%SZ)
+
+# The escalation RECORD is durable and unconditional; any notification built on
+# top of it is best-effort. Record and alert are deduplicated separately — a cap
+# on the alert alone lets the spool grow without bound exactly while the operator
+# is being told nothing, so the louder the failure the quieter it gets.
+#
+# A record field is arbitrary operator-supplied text: an identity comes from git
+# config or the environment, and cwd is whatever directory git ran in. Either can
+# contain a quote or a backslash, which would make the line unparseable JSON --
+# and a spool that silently stops parsing is worse than no spool, because the
+# operator only discovers it while reading the record that explains an outage.
+json_escape() {
+  local s=$1
+  s=${s//\\/\\\\}
+  s=${s//\"/\\\"}
+  s=${s//$'\t'/\\t}
+  s=${s//$'\r'/\\r}
+  s=${s//$'\n'/\\n}
+  printf '%s' "$s"
+}
+
+spool="${MOSAIC_CREDENTIAL_SPOOL:-$HOME/.local/state/mosaic-credential-escalations}"
+spool_record=""
+if mkdir -p "$spool" 2>/dev/null; then
+  chmod 700 "$spool" 2>/dev/null
+  spoolfile="$spool/$(date -u +%Y%m%d).jsonl"
+  dedupe="$spool/.spooled-${seat}-${ident:-none}-${reason}-$(date -u +%Y%m%d%H%M)"
+  if [ ! -e "$dedupe" ]; then
+    : > "$dedupe" 2>/dev/null
+    printf '{"ts":"%s","reason":"%s","identity":"%s","identity_source":"%s","kind":"%s","seat":"%s","host":"%s","cwd":"%s"}\n' \
+      "$(json_escape "$ts")" "$(json_escape "$reason")" \
+      "$(json_escape "${ident:-<unset>}")" "$(json_escape "$ident_src")" \
+      "$(json_escape "${ident_kind:-none}")" "$(json_escape "$seat")" \
+      "$(json_escape "$host")" "$(json_escape "$PWD")" \
+      >> "$spoolfile" 2>/dev/null
+    chmod 600 "$spoolfile" 2>/dev/null
+  fi
+  # Name the record only if one is actually on disk. Printing the path
+  # unconditionally sends the operator to a file that does not exist on exactly
+  # the hosts where the spool could not be created.
+  [ -s "$spoolfile" ] && spool_record="$spoolfile"
+  find "$spool" -maxdepth 1 -name '.spooled-*' -mmin +120 -delete 2>/dev/null
+fi
+
+cat >&2 <<EOF
+git-credential-mosaic: REFUSED (fail-closed).
+  host     : ${host}
+  identity : ${ident:-<unset>}${ident:+  (from ${ident_src}; resolved as a ${ident_kind})}
+  reason   : ${reason}
+EOF
+
+if [ -n "$ident" ]; then
+  cat >&2 <<EOF
+  expected : ${idtok}
+EOF
+fi
+
+cat >&2 <<EOF
+
+No per-identity credential resolved. This helper does NOT fall back to the shared
+account: that fallback makes every record it creates attributable to one
+principal, which is unrecoverable once a pull request has merged under it.
+
+Fix (pick one):
+  export MOSAIC_GIT_IDENTITY=<agent-id>        # process-scoped
+  git config mosaic.gitIdentity <agent-id>     # per-repo/worktree, persists
+Then provision that identity's credential at the path named above. An identity
+with a directory under \${MOSAIC_BRAIN_HOME:-\$HOME/.mosaic}/fleet/agents/ is a
+seat and is read ONLY from its own secrets/ slot; any other identity is read from
+~/.config/mosaic/secrets/gitea-tokens/. There is no fallback between the two.
+
+If this identity legitimately needs git access and has none, ask the orchestrator
+to provision one.
+
+EOF
+
+if [ -n "$spool_record" ]; then
+  echo "  record: ${spool_record}" >&2
+else
+  echo "  record: NOT WRITTEN — spool unavailable at ${spool}" >&2
+fi
+exit 1

packages/mosaic/framework/tools/git/test-git-credential-mosaic.sh

@@ -1,16 +1,26 @@
 #!/usr/bin/env bash
 # Regression harness for `git-credential-mosaic` — per-agent Gitea identity
-# resolution (Gate-16 author≠reviewer separation).
+# resolution (Gate-16 author≠reviewer separation) and fail-closed refusal.
 #
 # Covers:
 #   1. Identity resolution priority: MOSAIC_GIT_IDENTITY env > git config
 #      mosaic.gitIdentity (per-worktree) > git-supplied username.
-#   2. Correct per-slot token file path chosen per host
+#   2. Correct token file path chosen per host
 #      (gitea-usc-<id>.token vs gitea-mosaicstack-<id>.token).
-#   3. Per-slot token present -> emits that identity + token.
+#   3. Credential store selection: an identity with a directory under
-#   4. Per-slot token absent -> falls back to the shared account
+#      <brain>/fleet/agents/ is a SEAT and is read ONLY from its own secrets/
-#      (backward-compat / no-op for hosts without per-slot tokens).
+#      slot; any other identity is a SERVICE and is read from the framework
-#   5. Unknown/unrelated host -> exits 0 with no output (passthrough).
+#      store. No precedence between them and NO fallback from one to the other.
+#   4. Fail-closed: an identity that resolves but has no credential is REFUSED —
+#      no output, nonzero exit, a stderr diagnostic, and a durable spool record.
+#      The shared account is never emitted in its place.
+#   5. Fail-closed: no identity resolvable on a host that runs a fleet is also
+#      REFUSED, because records made there must name the agent that made them.
+#   6. Backward compatibility, the one surviving fallback: no identity AND no
+#      fleet -> shared account, unchanged. On such a host the shared account is
+#      the operator's own and there is no attribution to lose.
+#   7. Unknown/unrelated host -> exits 0 with no output (passthrough).
+#   8. Non-"get" verb -> exits 0 with no output.
 #
 # Uses stubbed token files under a fake HOME + a real (throwaway) git repo.
 # NEVER reads real secrets or touches the real ~/.config/mosaic/secrets.
@@ -21,6 +31,9 @@ SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 WORK_DIR="${MOSAIC_TEST_WORK_DIR:-$PWD/.mosaic-test-work/git-credential-mosaic}"
 FAKE_HOME="$WORK_DIR/home"
 REPO_DIR="$WORK_DIR/repo"
+BRAIN_DIR="$WORK_DIR/brain"
+SPOOL_DIR="$WORK_DIR/spool"
+SVC_STORE="$FAKE_HOME/.config/mosaic/secrets/gitea-tokens"
 # Mirror the real deployed layout (~/.config/mosaic/tools/{git,_lib}/) under the
 # fake HOME: git-credential-mosaic resolves its credentials.sh sibling via a
 # script-relative path (BASH_SOURCE), so the copy must live next to a stubbed
@@ -28,10 +41,10 @@ REPO_DIR="$WORK_DIR/repo"
 HELPER="$FAKE_HOME/.config/mosaic/tools/git/git-credential-mosaic"
 
 rm -rf "$WORK_DIR"
-mkdir -p "$FAKE_HOME/.config/mosaic/secrets/gitea-tokens" \
+mkdir -p "$SVC_STORE" \
          "$FAKE_HOME/.config/mosaic/tools/git" \
          "$FAKE_HOME/.config/mosaic/tools/_lib" \
-         "$REPO_DIR"
+         "$REPO_DIR" "$BRAIN_DIR"
 
 cp "$SCRIPT_DIR/git-credential-mosaic" "$HELPER"
 chmod +x "$HELPER"
@@ -68,7 +81,8 @@ run_helper() {
   local host="$1" username_in="$2"; shift 2
   (
     cd "$REPO_DIR"
-    env -i HOME="$FAKE_HOME" PATH="$PATH" "$@" bash "$HELPER" get <<EOF
+    env -i HOME="$FAKE_HOME" PATH="$PATH" MOSAIC_CREDENTIAL_SPOOL="$SPOOL_DIR" "$@" \
+      bash "$HELPER" get <<EOF
 host=$host
 username=$username_in
 
@@ -76,20 +90,61 @@ EOF
   )
 }
 
+# A refusal must be observable in four independent ways: nonzero exit, EMPTY
+# stdout, a stderr diagnostic naming the identity and host, and — the assertion
+# that actually catches a regression to the old behavior — NO shared token value
+# anywhere in the output. Checking only the exit code would pass against a helper
+# that emitted the shared credential and then exited 1.
+assert_fail_closed() {
+  local desc="$1" host="$2" username_in="$3" want_in_stderr="$4"; shift 4
+  local stderr_file="$WORK_DIR/stderr.tmp"
+  : > "$stderr_file"
+  set +e
+  local stdout
+  stdout=$(run_helper "$host" "$username_in" "$@" 2>"$stderr_file")
+  local rc=$?
+  set -e
+  local stderr
+  stderr=$(cat "$stderr_file")
+  if [[ "$rc" -eq 0 ]]; then
+    echo "FAIL: $desc — expected nonzero exit, got 0 (stdout='$stdout')" >&2
+    fail=1
+  fi
+  if [[ -n "$stdout" ]]; then
+    echo "FAIL: $desc — expected empty stdout (nothing emitted), got '$stdout'" >&2
+    fail=1
+  fi
+  if [[ "$stdout$stderr" == *"shared-mosaicstack-token"* || "$stdout$stderr" == *"shared-usc-token"* ]]; then
+    echo "FAIL: $desc — a SHARED token value appeared in the output. The shared-account fallback must be gone:" >&2
+    echo "$stdout$stderr" >&2
+    fail=1
+  fi
+  if [[ -n "$want_in_stderr" && "$stderr" != *"$want_in_stderr"* ]]; then
+    echo "FAIL: $desc — stderr does not contain '$want_in_stderr':" >&2
+    echo "$stderr" >&2
+    fail=1
+  fi
+  if [[ "$stderr" != *"$host"* ]]; then
+    echo "FAIL: $desc — stderr does not name the host '$host':" >&2
+    echo "$stderr" >&2
+    fail=1
+  fi
+}
+
 # ---------------------------------------------------------------------------
-# 1. No identity resolvable anywhere, no per-slot token -> shared fallback
+# 1. Backward compatibility: nothing resolvable, and NO fleet on this host ->
-#    (backward-compat: unchanged behavior when nothing is configured).
+#    shared account, unchanged. This is the only surviving fallback.
 # ---------------------------------------------------------------------------
 git -C "$REPO_DIR" config --unset mosaic.gitIdentity 2>/dev/null || true
 out=$(run_helper "git.mosaicstack.dev" "")
-assert_eq "shared fallback: username" "username=git" "$(echo "$out" | grep '^username=')"
+assert_eq "no identity + no fleet: username" "username=git" "$(echo "$out" | grep '^username=')"
-assert_eq "shared fallback: password" "password=shared-mosaicstack-token" "$(echo "$out" | grep '^password=')"
+assert_eq "no identity + no fleet: password" "password=shared-mosaicstack-token" "$(echo "$out" | grep '^password=')"
 
 # ---------------------------------------------------------------------------
-# 2. git-supplied username resolves to an identity WITH a per-slot token ->
+# 2. git-supplied username resolves to a SERVICE identity WITH a token in the
-#    that identity + token wins over the shared account.
+#    framework store -> that identity + token wins over the shared account.
 # ---------------------------------------------------------------------------
-echo -n "agentA-mosaicstack-token" > "$FAKE_HOME/.config/mosaic/secrets/gitea-tokens/gitea-mosaicstack-agentA.token"
+echo -n "agentA-mosaicstack-token" > "$SVC_STORE/gitea-mosaicstack-agentA.token"
 out=$(run_helper "git.mosaicstack.dev" "agentA")
 assert_eq "username-resolved identity: username" "username=agentA" "$(echo "$out" | grep '^username=')"
 assert_eq "username-resolved identity: password" "password=agentA-mosaicstack-token" "$(echo "$out" | grep '^password=')"
@@ -97,7 +152,7 @@ assert_eq "username-resolved identity: password" "password=agentA-mosaicstack-to
 # ---------------------------------------------------------------------------
 # 3. git config mosaic.gitIdentity (per-worktree) beats git-supplied username.
 # ---------------------------------------------------------------------------
-echo -n "agentB-mosaicstack-token" > "$FAKE_HOME/.config/mosaic/secrets/gitea-tokens/gitea-mosaicstack-agentB.token"
+echo -n "agentB-mosaicstack-token" > "$SVC_STORE/gitea-mosaicstack-agentB.token"
 git -C "$REPO_DIR" config mosaic.gitIdentity agentB
 out=$(run_helper "git.mosaicstack.dev" "agentA")
 assert_eq "git-config beats username: username" "username=agentB" "$(echo "$out" | grep '^username=')"
@@ -106,54 +161,210 @@ assert_eq "git-config beats username: password" "password=agentB-mosaicstack-tok
 # ---------------------------------------------------------------------------
 # 4. MOSAIC_GIT_IDENTITY env beats git config mosaic.gitIdentity.
 # ---------------------------------------------------------------------------
-echo -n "agentC-mosaicstack-token" > "$FAKE_HOME/.config/mosaic/secrets/gitea-tokens/gitea-mosaicstack-agentC.token"
+echo -n "agentC-mosaicstack-token" > "$SVC_STORE/gitea-mosaicstack-agentC.token"
 out=$(run_helper "git.mosaicstack.dev" "agentA" MOSAIC_GIT_IDENTITY=agentC)
 assert_eq "env beats git-config: username" "username=agentC" "$(echo "$out" | grep '^username=')"
 assert_eq "env beats git-config: password" "password=agentC-mosaicstack-token" "$(echo "$out" | grep '^password=')"
 git -C "$REPO_DIR" config --unset mosaic.gitIdentity
 
 # ---------------------------------------------------------------------------
-# 5. Identity resolves, but no matching per-slot token file -> falls back to
+# 5. Correct token PATH is chosen per host: same agent id, different host
-#    the shared account (per-agent identity is opt-in, not a hard requirement).
+#    prefix (gitea-usc- vs gitea-mosaicstack-).
 # ---------------------------------------------------------------------------
-out=$(run_helper "git.mosaicstack.dev" "no-such-agent")
+echo -n "agentD-usc-token" > "$SVC_STORE/gitea-usc-agentD.token"
-assert_eq "no per-slot token: username" "username=git" "$(echo "$out" | grep '^username=')"
-assert_eq "no per-slot token: password" "password=shared-mosaicstack-token" "$(echo "$out" | grep '^password=')"
-
-# ---------------------------------------------------------------------------
-# 6. Correct per-slot token PATH is chosen per host: same agent id, different
-#    host prefix (gitea-usc- vs gitea-mosaicstack-).
-# ---------------------------------------------------------------------------
-echo -n "agentD-usc-token" > "$FAKE_HOME/.config/mosaic/secrets/gitea-tokens/gitea-usc-agentD.token"
 out=$(run_helper "git.uscllc.com" "agentD")
 assert_eq "host-scoped token path (usc): username" "username=agentD" "$(echo "$out" | grep '^username=')"
 assert_eq "host-scoped token path (usc): password" "password=agentD-usc-token" "$(echo "$out" | grep '^password=')"
-# agentD has NO mosaicstack token -> must fall back to shared mosaicstack, not
-# leak the usc token across hosts.
-out=$(run_helper "git.mosaicstack.dev" "agentD")
-assert_eq "host-scoped token path (cross-host must not leak): username" "username=git" "$(echo "$out" | grep '^username=')"
-assert_eq "host-scoped token path (cross-host must not leak): password" "password=shared-mosaicstack-token" "$(echo "$out" | grep '^password=')"
 
 # ---------------------------------------------------------------------------
-# 7. Unrelated/unknown host -> exit 0, no output (passthrough for non-Gitea
+# 6. FAIL CLOSED — identity resolves, no credential for it on this host. Must
-#    remotes, e.g. github.com via a different credential helper).
+#    NOT borrow the shared account, and must NOT leak the same agent's token
+#    for a DIFFERENT host (agentD holds a usc token and no mosaicstack one).
+# ---------------------------------------------------------------------------
+assert_fail_closed "cross-host absence refuses (no shared fallback, no cross-host leak)" \
+  "git.mosaicstack.dev" "agentD" "gitea-mosaicstack-agentD.token"
+# The agent's own usc token must not appear either.
+: > "$WORK_DIR/stderr.tmp"
+set +e
+leak_out=$(run_helper "git.mosaicstack.dev" "agentD" 2>"$WORK_DIR/stderr.tmp")
+set -e
+if [[ "$leak_out$(cat "$WORK_DIR/stderr.tmp")" == *"agentD-usc-token"* ]]; then
+  echo "FAIL: cross-host leak — the usc token value appeared on a mosaicstack request" >&2
+  fail=1
+fi
+
+assert_fail_closed "unknown identity refuses (shared account never substituted)" \
+  "git.mosaicstack.dev" "no-such-agent" "no-token-for-identity"
+
+# ---------------------------------------------------------------------------
+# 7. A refusal leaves a durable spool record, and that record contains no token.
+#    The stderr diagnostic is transient; the record is what an operator reads
+#    afterwards, so it must exist independently of anyone watching the terminal.
+# ---------------------------------------------------------------------------
+spool_file=$(find "$SPOOL_DIR" -maxdepth 1 -name '*.jsonl' | head -n 1)
+if [[ -z "$spool_file" ]]; then
+  echo "FAIL: fail-closed left no spool record under $SPOOL_DIR" >&2
+  fail=1
+else
+  spool_body=$(cat "$spool_file")
+  assert_eq "spool record names the refused identity" "1" \
+    "$(grep -c '"identity":"no-such-agent"' "$spool_file" | head -n 1)"
+  if [[ "$spool_body" == *"shared-"*"-token"* || "$spool_body" == *"agentD-usc-token"* ]]; then
+    echo "FAIL: spool record contains a token value:" >&2
+    echo "$spool_body" >&2
+    fail=1
+  fi
+fi
+
+# ---------------------------------------------------------------------------
+# 8. SEAT identity: an id with a directory under <brain>/fleet/agents/ is read
+#    from its OWN secrets/ slot, not from the framework store.
+# ---------------------------------------------------------------------------
+mkdir -p "$BRAIN_DIR/fleet/agents/seatE/secrets"
+echo -n "seatE-slot-token" > "$BRAIN_DIR/fleet/agents/seatE/secrets/gitea-mosaicstack-seatE.token"
+out=$(run_helper "git.mosaicstack.dev" "seatE" MOSAIC_BRAIN_HOME="$BRAIN_DIR")
+assert_eq "seat reads its own slot: username" "username=seatE" "$(echo "$out" | grep '^username=')"
+assert_eq "seat reads its own slot: password" "password=seatE-slot-token" "$(echo "$out" | grep '^password=')"
+
+# ---------------------------------------------------------------------------
+# 9. NO CROSS-STORE FALLBACK — the assertion this whole store-selection design
+#    exists for. seatF is a seat (it has a directory) with an EMPTY slot, while
+#    a framework-store token of the identical name is present and readable.
+#    The helper must refuse rather than read it: one credential, one location,
+#    and a seat that reads a same-named service credential is exactly the
+#    silent-substitution failure the fail-closed rule removes.
+# ---------------------------------------------------------------------------
+mkdir -p "$BRAIN_DIR/fleet/agents/seatF/secrets"
+echo -n "seatF-SERVICE-STORE-token" > "$SVC_STORE/gitea-mosaicstack-seatF.token"
+assert_fail_closed "seat with empty slot does NOT fall back to the framework store" \
+  "git.mosaicstack.dev" "seatF" "fleet/agents/seatF/secrets" MOSAIC_BRAIN_HOME="$BRAIN_DIR"
+: > "$WORK_DIR/stderr.tmp"
+set +e
+xstore_out=$(run_helper "git.mosaicstack.dev" "seatF" MOSAIC_BRAIN_HOME="$BRAIN_DIR" 2>"$WORK_DIR/stderr.tmp")
+set -e
+if [[ "$xstore_out$(cat "$WORK_DIR/stderr.tmp")" == *"seatF-SERVICE-STORE-token"* ]]; then
+  echo "FAIL: cross-store fallback — a seat read the framework store's same-named token" >&2
+  fail=1
+fi
+# Control: that framework-store token IS readable, so the refusal above is the
+# store rule firing and not an unreadable file. A non-seat identity pointed at
+# the same file gets it.
+out=$(run_helper "git.mosaicstack.dev" "seatF" MOSAIC_BRAIN_HOME="$WORK_DIR/no-such-brain")
+assert_eq "control — same file IS readable for a non-seat identity" \
+  "password=seatF-SERVICE-STORE-token" "$(echo "$out" | grep '^password=')"
+
+# ---------------------------------------------------------------------------
+# 10. FAIL CLOSED — no identity resolvable, but this host runs a fleet. Where
+#     seats exist, an unattributable request is refused instead of receiving
+#     the shared account. Contrast with case 1, which is the same request on a
+#     host with no fleet and still returns the shared account.
+# ---------------------------------------------------------------------------
+git -C "$REPO_DIR" config --unset mosaic.gitIdentity 2>/dev/null || true
+assert_fail_closed "no identity on a fleet host refuses" \
+  "git.mosaicstack.dev" "" "no-identity" MOSAIC_BRAIN_HOME="$BRAIN_DIR"
+
+# ---------------------------------------------------------------------------
+# 11. The brain home defaults to ~/.mosaic when MOSAIC_BRAIN_HOME is unset —
+#     the fleet gate must fire on the default path too, not only on an
+#     explicitly injected one. Case 1 ran before this directory existed; the
+#     same call now refuses, which also proves case 1 was measuring the
+#     no-fleet branch rather than passing for an unrelated reason.
+# ---------------------------------------------------------------------------
+mkdir -p "$FAKE_HOME/.mosaic/fleet/agents"
+assert_fail_closed "fleet gate fires on the default ~/.mosaic brain home" \
+  "git.mosaicstack.dev" "" "no-identity"
+
+# ---------------------------------------------------------------------------
+# 12. Unrelated/unknown host -> exit 0, no output (passthrough for non-Gitea
+#     remotes, e.g. github.com via a different credential helper). A fleet host
+#     must not refuse a host this helper does not own.
 # ---------------------------------------------------------------------------
 out=$(run_helper "github.com" "agentA")
 assert_eq "unknown host: no output" "" "$out"
+out=$(run_helper "github.com" "" MOSAIC_BRAIN_HOME="$BRAIN_DIR")
+assert_eq "unknown host on a fleet host: still passthrough, not a refusal" "" "$out"
 
 # ---------------------------------------------------------------------------
-# 8. Non-"get" verb (store/erase) -> exit 0, no output (git-credential
+# 13. Non-"get" verb (store/erase) -> exit 0, no output (git-credential
-#    protocol: this helper only implements get).
+#     protocol: this helper only implements get).
 # ---------------------------------------------------------------------------
 store_out=$(cd "$REPO_DIR" && env -i HOME="$FAKE_HOME" PATH="$PATH" bash "$HELPER" store <<EOF
 host=git.mosaicstack.dev
-username=agentA
+username=no-such-agent
 password=whatever
 
 EOF
 )
 assert_eq "store verb: no output" "" "$store_out"
 
+# ---------------------------------------------------------------------------
+# 14. The escalation record is machine-readable even when a field carries a
+#     quote or a backslash. A cwd is arbitrary operator text; an unescaped one
+#     silently turns the spool into unparseable JSONL, and the operator only
+#     finds out while reading the record that explains an outage.
+# ---------------------------------------------------------------------------
+hostile_dir="$WORK_DIR/we\"ird\\dir"
+mkdir -p "$hostile_dir"
+hostile_spool="$WORK_DIR/spool-hostile"
+(
+  cd "$hostile_dir"
+  env -i HOME="$FAKE_HOME" PATH="$PATH" MOSAIC_CREDENTIAL_SPOOL="$hostile_spool" \
+    MOSAIC_GIT_IDENTITY=no-such-agent \
+    bash "$HELPER" get <<EOF >/dev/null 2>&1
+host=git.mosaicstack.dev
+username=no-such-agent
+
+EOF
+) || true
+# Deliberately not `ls ... | head -1`: under `set -o pipefail` a missed glob
+# makes ls exit 2, the pipeline inherits it, and `set -e` kills this suite with
+# zero output — the same silent-nonzero failure this file exists to catch.
+record_file=""
+for candidate in "$hostile_spool"/*.jsonl; do
+  if [[ -e "$candidate" ]]; then
+    record_file="$candidate"
+    break
+  fi
+done
+if [[ -z "$record_file" ]]; then
+  echo "FAIL: hostile cwd — no escalation record was written at all" >&2
+  fail=1
+elif ! python3 -c 'import json,sys
+for line in open(sys.argv[1]):
+    line = line.strip()
+    if line:
+        json.loads(line)' "$record_file" 2>/dev/null; then
+  echo "FAIL: hostile cwd — escalation record is not parseable JSONL:" >&2
+  cat "$record_file" >&2
+  fail=1
+fi
+
+# ---------------------------------------------------------------------------
+# 15. When the spool cannot be created, the diagnostic must NOT name a record
+#     path. Naming a file that was never written sends the operator to an
+#     empty path on exactly the hosts where the escalation was lost.
+# ---------------------------------------------------------------------------
+unwritable_spool="/proc/mosaic-credential-spool-cannot-exist"
+nospool_err=$(
+  cd "$REPO_DIR"
+  env -i HOME="$FAKE_HOME" PATH="$PATH" MOSAIC_CREDENTIAL_SPOOL="$unwritable_spool" \
+    MOSAIC_GIT_IDENTITY=no-such-agent \
+    bash "$HELPER" get <<EOF 2>&1 >/dev/null
+host=git.mosaicstack.dev
+username=no-such-agent
+
+EOF
+) || true
+if [[ "$nospool_err" == *"record: $unwritable_spool/"* ]]; then
+  echo "FAIL: unwritable spool — diagnostic names a record file that was never written" >&2
+  fail=1
+fi
+if [[ "$nospool_err" != *"NOT WRITTEN"* ]]; then
+  echo "FAIL: unwritable spool — diagnostic does not say the record was not written" >&2
+  echo "$nospool_err" >&2
+  fail=1
+fi
+
 if [[ "$fail" -eq 0 ]]; then
   echo "git-credential-mosaic identity resolution regression passed"
 fi

packages/mosaic/framework/tools/git/test-gitea-token-identity.sh

@@ -23,6 +23,17 @@
 #   6. Scope containment: identity requested + an UNRECOGNIZED Gitea host (no
 #      per-slot token scheme) -> Patch 2b does not apply; existing
 #      fall-through behavior is unchanged.
+#   7. Credential store selection: an identity with a directory under
+#      <brain>/fleet/agents/ is a SEAT and is read ONLY from its own secrets/
+#      slot; any other identity is a SERVICE and is read from the framework
+#      store. No precedence between them and NO fallback from one to the
+#      other — a seat with an empty slot is REFUSED even when a same-named
+#      token sits in the framework store.
+#   8. Fail loud when NO identity resolves on a host that runs a fleet: where
+#      seats exist, an unattributable API call is refused rather than made
+#      under the shared account. On a host with no fleet the same call still
+#      returns the shared token (case 1), which is what keeps this change a
+#      no-op for non-fleet operators of the framework.
 #
 # Uses a stubbed credentials.json + stubbed per-slot token files under a fake
 # HOME. NEVER reads real secrets or touches the real ~/.config/mosaic/secrets.
@@ -210,6 +221,94 @@ if [[ "$err" == *"no per-slot token at"* ]]; then
 fi
 git -C "$REPO_DIR" config --unset mosaic.gitIdentity
 
+# ---------------------------------------------------------------------------
+# 7. SEAT identity: an id with a directory under <brain>/fleet/agents/ is read
+#    from its OWN secrets/ slot, not from the framework store. The brain home
+#    is resolved exactly as packages/mosaic/src/fleet/brain-home.ts does it:
+#    MOSAIC_BRAIN_HOME, else ~/.mosaic.
+# ---------------------------------------------------------------------------
+BRAIN_DIR="$WORK_DIR/brain"
+mkdir -p "$BRAIN_DIR/fleet/agents/seatE/secrets"
+echo -n "seatE-slot-token" > "$BRAIN_DIR/fleet/agents/seatE/secrets/gitea-mosaicstack-seatE.token"
+out=$(call_get_gitea_token "git.mosaicstack.dev" MOSAIC_GIT_IDENTITY=seatE MOSAIC_BRAIN_HOME="$BRAIN_DIR")
+assert_eq "seat reads its own slot" "seatE-slot-token" "$out"
+
+# ---------------------------------------------------------------------------
+# 8. NO CROSS-STORE FALLBACK. seatF is a seat (it has a directory) with an
+#    EMPTY slot, while a framework-store token of the identical name is
+#    present and readable. It must be REFUSED rather than served that token:
+#    one credential, one location. A seat that silently reads a same-named
+#    service credential is the same substitution failure as the shared-account
+#    fallback, one store further down.
+# ---------------------------------------------------------------------------
+mkdir -p "$BRAIN_DIR/fleet/agents/seatF/secrets"
+echo -n "seatF-SERVICE-STORE-token" > "$FAKE_HOME/.config/mosaic/secrets/gitea-tokens/gitea-mosaicstack-seatF.token"
+assert_failloud "seat with empty slot does NOT fall back to the framework store" \
+  "git.mosaicstack.dev" "seatF" \
+  "$BRAIN_DIR/fleet/agents/seatF/secrets/gitea-mosaicstack-seatF.token" \
+  MOSAIC_GIT_IDENTITY=seatF MOSAIC_BRAIN_HOME="$BRAIN_DIR"
+# assert_failloud only screens stderr for the word "shared"; this store's token
+# is not named that, so check for its value explicitly.
+set +e
+xstore_out=$(call_get_gitea_token "git.mosaicstack.dev" MOSAIC_GIT_IDENTITY=seatF MOSAIC_BRAIN_HOME="$BRAIN_DIR" 2>"$WORK_DIR/stderr-xstore.tmp")
+set -e
+if [[ "$xstore_out$(cat "$WORK_DIR/stderr-xstore.tmp")" == *"seatF-SERVICE-STORE-token"* ]]; then
+  echo "FAIL: cross-store fallback — a seat was served the framework store's same-named token" >&2
+  fail=1
+fi
+# Control: that framework-store token IS readable, so the refusal above is the
+# store rule firing and not an unreadable file. The same id, resolved against a
+# brain home where it is not a seat, gets it.
+out=$(call_get_gitea_token "git.mosaicstack.dev" MOSAIC_GIT_IDENTITY=seatF MOSAIC_BRAIN_HOME="$WORK_DIR/no-such-brain")
+assert_eq "control — same file IS readable for a non-seat identity" "seatF-SERVICE-STORE-token" "$out"
+
+# ---------------------------------------------------------------------------
+# 9. FAIL LOUD — no identity resolvable, but this host runs a fleet. Contrast
+#    with case 1: the identical call on a host with no fleet still returns the
+#    shared token.
+# ---------------------------------------------------------------------------
+git -C "$REPO_DIR" config --unset mosaic.gitIdentity 2>/dev/null || true
+set +e
+noid_out=$(call_get_gitea_token "git.mosaicstack.dev" MOSAIC_BRAIN_HOME="$BRAIN_DIR" 2>"$WORK_DIR/stderr-noid.tmp")
+noid_rc=$?
+set -e
+noid_err=$(cat "$WORK_DIR/stderr-noid.tmp")
+if [[ "$noid_rc" -eq 0 ]]; then
+  echo "FAIL: no identity on a fleet host — expected nonzero return, got 0 (stdout='$noid_out')" >&2
+  fail=1
+fi
+if [[ -n "$noid_out" ]]; then
+  echo "FAIL: no identity on a fleet host — expected empty stdout, got '$noid_out'" >&2
+  fail=1
+fi
+if [[ "$noid_out" == *"shared-mosaicstack-token"* || "$noid_err" == *"shared-mosaicstack-token"* ]]; then
+  echo "FAIL: no identity on a fleet host — the shared token was served anyway:" >&2
+  echo "$noid_out$noid_err" >&2
+  fail=1
+fi
+if [[ "$noid_err" != *"MOSAIC_GIT_IDENTITY"* ]]; then
+  echo "FAIL: no identity on a fleet host — stderr does not say how to set an identity:" >&2
+  echo "$noid_err" >&2
+  fail=1
+fi
+
+# ---------------------------------------------------------------------------
+# 10. The brain home defaults to ~/.mosaic when MOSAIC_BRAIN_HOME is unset —
+#     the fleet gate must fire on the default path too, not only on an
+#     explicitly injected one. Case 1 ran before this directory existed; the
+#     same call now refuses, which also proves case 1 was measuring the
+#     no-fleet branch rather than passing for an unrelated reason.
+# ---------------------------------------------------------------------------
+mkdir -p "$FAKE_HOME/.mosaic/fleet/agents"
+set +e
+dflt_out=$(call_get_gitea_token "git.mosaicstack.dev" 2>"$WORK_DIR/stderr-dflt.tmp")
+dflt_rc=$?
+set -e
+if [[ "$dflt_rc" -eq 0 || -n "$dflt_out" ]]; then
+  echo "FAIL: fleet gate did not fire on the default ~/.mosaic brain home (rc=$dflt_rc stdout='$dflt_out')" >&2
+  fail=1
+fi
+
 if [[ "$fail" -eq 0 ]]; then
   echo "get_gitea_token identity resolution regression passed"
 fi

packages/mosaic/src/commands/mission-prd.spec.ts

@@ -0,0 +1,149 @@
+import { mkdtemp, readFile, readdir } from 'node:fs/promises';
+import os from 'node:os';
+import path from 'node:path';
+
+import { parse as parseYaml } from 'yaml';
+import { Command } from 'commander';
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+
+import { registerMissionCommand } from './mission.js';
+import { PrdService } from '@mosaicstack/prdy';
+import type { MissionInfo } from '../tui/gateway-api.js';
+
+// ── Mocks: the gateway is not available in adapter tests ──────────────────────
+
+// vi.hoisted: the mock factory is hoisted above imports, so the fixture must
+// be initialized there too.
+const MISSION = vi.hoisted(
+  (): MissionInfo => ({
+    id: 'mission-plan-1',
+    name: 'Plan Mission Alpha',
+    description: null,
+    status: 'planning',
+    projectId: null,
+    userId: null,
+    phase: null,
+    milestones: null,
+    config: null,
+    createdAt: '2026-01-01T00:00:00.000Z',
+    updatedAt: '2026-03-04T05:06:07.000Z',
+  }),
+);
+
+vi.mock('./with-auth.js', () => ({
+  withAuth: vi.fn().mockResolvedValue({
+    gateway: 'http://localhost:14242',
+    cookie: 'better-auth.session_token=test',
+    session: {},
+  }),
+}));
+
+vi.mock('../tui/gateway-api.js', () => ({
+  fetchMissions: vi.fn().mockResolvedValue([MISSION]),
+  fetchMission: vi.fn(),
+  createMission: vi.fn(),
+  updateMission: vi.fn(),
+  fetchMissionTasks: vi.fn().mockResolvedValue([]),
+  createMissionTask: vi.fn(),
+  updateMissionTask: vi.fn(),
+  fetchProjects: vi.fn().mockResolvedValue([]),
+}));
+
+// ── Helpers ──────────────────────────────────────────────────────────────────
+
+const originalCwd = process.cwd();
+let projectDir: string;
+let logSpy: ReturnType<typeof vi.spyOn>;
+let consoleStub: ReturnType<typeof vi.spyOn>[] = [];
+
+function buildTestProgram(): Command {
+  const program = new Command('mosaic').exitOverride();
+  registerMissionCommand(program);
+  return program;
+}
+
+beforeEach(async () => {
+  projectDir = await mkdtemp(path.join(os.tmpdir(), 'mosaic-mission-plan-'));
+  process.chdir(projectDir);
+  logSpy = vi.spyOn(console, 'log').mockImplementation(() => {});
+  consoleStub.push(logSpy);
+});
+
+afterEach(() => {
+  // Restore only the per-test spies; module factory mocks keep their
+  // implementations across tests.
+  for (const stub of consoleStub) stub.mockRestore();
+  consoleStub = [];
+  process.chdir(originalCwd);
+});
+
+// ── Tests ────────────────────────────────────────────────────────────────────
+
+describe('mosaic mission --plan (thin adapter over PrdService)', () => {
+  it('creates the PRD in the shared docs/prdy authority store and persists the mission linkage', async () => {
+    await buildTestProgram().parseAsync(['mission', '--plan', 'Plan Mission Alpha'], {
+      from: 'user',
+    });
+
+    // PRD landed in the same store `mosaic prdy` uses.
+    const files = await readdir(path.join(projectDir, 'docs', 'prdy'));
+    expect(files).toHaveLength(1);
+    expect(files[0]).toMatch(/\.yaml$/);
+
+    // Fresh service instance (new-process equivalent) reads the linkage back.
+    const service = new PrdService({ projectPath: projectDir });
+    const docs = await service.list();
+    expect(docs).toHaveLength(1);
+
+    const prd = docs[0]!;
+    expect(prd.title).toBe('Plan Mission Alpha');
+    expect(prd.version).toBe(1);
+
+    const links = await service.listMissionLinks(prd.id);
+    expect(links).toHaveLength(1);
+    expect(links[0]).toMatchObject({
+      missionId: MISSION.id,
+      missionVersion: MISSION.updatedAt, // mission version marker
+      prdVersion: 1,
+    });
+
+    expect(logSpy).toHaveBeenCalledWith(expect.stringContaining('PRD created and linked'));
+  });
+
+  it('linkage is persisted in the YAML authority document itself (survives restart)', async () => {
+    await buildTestProgram().parseAsync(['mission', '--plan', 'Plan Mission Alpha'], {
+      from: 'user',
+    });
+
+    const files = await readdir(path.join(projectDir, 'docs', 'prdy'));
+    const raw = await readFile(path.join(projectDir, 'docs', 'prdy', files[0]!), 'utf8');
+    const persisted = parseYaml(raw) as { missions: Array<Record<string, unknown>> };
+
+    expect(persisted.missions).toHaveLength(1);
+    expect(persisted.missions[0]).toMatchObject({ missionId: 'mission-plan-1' });
+  });
+
+  it('the mission path and the prdy path resolve to the same store with stable ids/versions', async () => {
+    // Mission path.
+    await buildTestProgram().parseAsync(['mission', '--plan', 'Plan Mission Alpha'], {
+      from: 'user',
+    });
+
+    // prdy path (service, non-interactive entry).
+    const service = new PrdService({ projectPath: projectDir });
+    const direct = await service.create({ name: 'Directly Created' });
+
+    const all = await service.list();
+    expect(all.map((doc) => doc.id).sort()).toEqual([...all.map((doc) => doc.id)].sort());
+    expect(all).toHaveLength(2);
+
+    const files = await readdir(path.join(projectDir, 'docs', 'prdy'));
+    expect(files).toContain(`${direct.id}.yaml`);
+
+    // Both are v1 in the same store with distinct stable ids.
+    for (const doc of all) {
+      expect(doc.version).toBe(1);
+      expect(files).toContain(`${doc.id}.yaml`);
+    }
+  });
+});

packages/mosaic/src/commands/mission.ts

@@ -256,14 +256,41 @@ async function planMission(
   console.log(`Planning mission: ${mission.name}\n`);
 
   try {
-    const { runPrdWizard } = await import('@mosaicstack/prdy');
+    // Thin adapter: the PRD authority (create + mission↔PRD linkage) lives in
-    await runPrdWizard({
+    // PrdService — no second writer path. The mission's updatedAt serves as
+    // its version marker (the gateway exposes no numeric mission version).
+    const { PrdService, runPrdWizard } = await import('@mosaicstack/prdy');
+    const service = new PrdService({ projectPath: process.cwd() });
+
+    if (process.stdout.isTTY) {
+      const created = await runPrdWizard({
+        name: mission.name,
+        projectPath: process.cwd(),
+        interactive: true,
+      });
+      const linked = await service.linkMission({
+        prdId: created.id,
+        missionId: mission.id,
+        missionVersion: mission.updatedAt,
+        requirementIds: [],
+      });
+      console.log(
+        `\nMission ${mission.id} linked to PRD ${linked.id} v${linked.version} (docs/prdy/).`,
+      );
+      return;
+    }
+
+    const doc = await service.planForMission({
       name: mission.name,
-      projectPath: process.cwd(),
+      missionId: mission.id,
-      interactive: true,
+      missionVersion: mission.updatedAt,
+      requirementIds: [],
     });
+    console.log(
+      `PRD created and linked: ${doc.id} v${doc.version} — mission ${mission.id} (docs/prdy/).`,
+    );
   } catch (err) {
-    console.error(`PRD wizard failed: ${err instanceof Error ? err.message : String(err)}`);
+    console.error(`PRD planning failed: ${err instanceof Error ? err.message : String(err)}`);
     process.exit(1);
   }
 }

packages/mosaic/src/commands/prdy.spec.ts

@@ -0,0 +1,204 @@
+import { mkdtemp, readFile, readdir, writeFile } from 'node:fs/promises';
+import os from 'node:os';
+import path from 'node:path';
+
+import { stringify as stringifyYaml } from 'yaml';
+import { Command } from 'commander';
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+
+import { registerPrdyCommand } from './prdy.js';
+import { PrdService } from '@mosaicstack/prdy';
+
+// ── Mocks: keep the adapter test offline (no gateway, no disk side effects
+//    outside the tmp project dir) ──────────────────────────────────────────────
+
+vi.mock('./with-auth.js', () => ({
+  withAuth: vi.fn().mockResolvedValue({
+    gateway: 'http://localhost:14242',
+    cookie: 'better-auth.session_token=test',
+    session: {},
+  }),
+}));
+
+vi.mock('../tui/gateway-api.js', () => ({
+  fetchProjects: vi.fn().mockResolvedValue([]),
+}));
+
+// ── Helpers ──────────────────────────────────────────────────────────────────
+
+class ProcessExitError extends Error {
+  constructor(readonly code: number) {
+    super(`process.exit(${code})`);
+  }
+}
+
+function stubProcessExit() {
+  return vi.spyOn(process, 'exit').mockImplementation(((code?: number) => {
+    throw new ProcessExitError(code ?? 0);
+  }) as never);
+}
+
+const originalCwd = process.cwd();
+let projectDir: string;
+let errorSpy: ReturnType<typeof vi.spyOn>;
+let logSpy: ReturnType<typeof vi.spyOn>;
+let exitStub: ReturnType<typeof stubProcessExit>;
+
+function buildTestProgram(): Command {
+  const program = new Command('mosaic').exitOverride();
+  registerPrdyCommand(program);
+  return program;
+}
+
+function runPrdy(args: string[]): Promise<unknown> {
+  return buildTestProgram().parseAsync(['prdy', ...args], { from: 'user' });
+}
+
+function importableDocument(overrides: Record<string, unknown> = {}): Record<string, unknown> {
+  return {
+    id: 'cmd-import-prd',
+    title: 'Command Import PRD',
+    status: 'approved', // must be forced to draft: validity is not approval
+    projectPath: '/tmp/elsewhere',
+    template: 'software',
+    version: 1,
+    sections: [
+      { id: 'introduction', title: 'Introduction', fields: { context: 'x', objective: 'y' } },
+    ],
+    missions: [],
+    createdAt: '2026-01-01T00:00:00.000Z',
+    updatedAt: '2026-01-01T00:00:00.000Z',
+    ...overrides,
+  };
+}
+
+beforeEach(async () => {
+  projectDir = await mkdtemp(path.join(os.tmpdir(), 'mosaic-prdy-'));
+  process.chdir(projectDir);
+  exitStub = stubProcessExit();
+  errorSpy = vi.spyOn(console, 'error').mockImplementation(() => {});
+  logSpy = vi.spyOn(console, 'log').mockImplementation(() => {});
+});
+
+afterEach(() => {
+  // Restore only the per-test spies: module factory mocks must keep their
+  // implementations for the next test.
+  exitStub.mockRestore();
+  errorSpy.mockRestore();
+  logSpy.mockRestore();
+  process.chdir(originalCwd);
+});
+
+// ── Tests ────────────────────────────────────────────────────────────────────
+
+describe('mosaic prdy (thin adapter over PrdService)', () => {
+  it('non-interactive --init creates a PRD in the docs/prdy authority store', async () => {
+    await runPrdy(['--init', 'Adapter Created']);
+
+    const files = await readdir(path.join(projectDir, 'docs', 'prdy'));
+    expect(files).toHaveLength(1);
+    expect(files[0]).toMatch(/\.yaml$/);
+
+    const docs = await new PrdService({ projectPath: projectDir }).list();
+    expect(docs).toHaveLength(1);
+    expect(docs[0]?.title).toBe('Adapter Created');
+    expect(docs[0]?.version).toBe(1);
+    expect(logSpy).toHaveBeenCalledWith(expect.stringContaining('PRD created'));
+  });
+
+  it('--import <file> creates a valid import through the service', async () => {
+    const filePath = path.join(projectDir, 'incoming.yaml');
+    await writeFile(filePath, stringifyYaml(importableDocument()), 'utf8');
+
+    await runPrdy(['--import', filePath]);
+
+    const docs = await new PrdService({ projectPath: projectDir }).list();
+    expect(docs).toHaveLength(1);
+    expect(docs[0]?.id).toBe('cmd-import-prd');
+    expect(docs[0]?.status).toBe('draft'); // import ≠ approval
+    expect(logSpy).toHaveBeenCalledWith(expect.stringContaining('Imported PRD cmd-import-prd'));
+  });
+
+  it('--import of a structurally-invalid file is a typed refusal that creates nothing', async () => {
+    const filePath = path.join(projectDir, 'broken.yaml');
+    await writeFile(filePath, stringifyYaml({ id: 'incomplete', no: 'structure' }), 'utf8');
+
+    await expect(runPrdy(['--import', filePath])).rejects.toBeInstanceOf(ProcessExitError);
+
+    // Typed refusal surfaced to the user, nothing created.
+    expect(errorSpy).toHaveBeenCalledWith(expect.stringContaining('PRD wizard failed'));
+    await expect(readdir(path.join(projectDir, 'docs'))).rejects.toMatchObject({ code: 'ENOENT' });
+  });
+
+  it('--import on conflict refuses with a successor proposal and leaves bytes untouched', async () => {
+    const service = new PrdService({ projectPath: projectDir });
+    const existing = await service.create({ name: 'Conflict Target' });
+    const storeFile = path.join(projectDir, 'docs', 'prdy', `${existing.id}.yaml`);
+    const beforeBytes = await readFile(storeFile, 'utf8');
+
+    const filePath = path.join(projectDir, 'divergent.yaml');
+    await writeFile(
+      filePath,
+      stringifyYaml(
+        importableDocument({
+          ...existing,
+          title: 'Divergent Command Import',
+        }),
+      ),
+      'utf8',
+    );
+
+    await expect(runPrdy(['--import', filePath])).rejects.toBeInstanceOf(ProcessExitError);
+
+    expect(errorSpy).toHaveBeenCalledWith(expect.stringContaining('refusing to overwrite'));
+    expect(errorSpy).toHaveBeenCalledWith(expect.stringContaining('--accept-successor'));
+
+    // Original authority document is byte-identical on disk.
+    expect(await readFile(storeFile, 'utf8')).toBe(beforeBytes);
+  });
+
+  it('--import --accept-successor persists the successor version explicitly', async () => {
+    const service = new PrdService({ projectPath: projectDir });
+    const existing = await service.create({ name: 'Successor Target' });
+
+    const filePath = path.join(projectDir, 'divergent2.yaml');
+    await writeFile(
+      filePath,
+      stringifyYaml(
+        importableDocument({
+          ...existing,
+          title: 'Accepted Via CLI',
+        }),
+      ),
+      'utf8',
+    );
+
+    await runPrdy(['--import', filePath, '--accept-successor']);
+
+    const doc = await service.get(existing.id);
+    expect(doc.version).toBe(2);
+    expect(doc.title).toBe('Accepted Via CLI');
+    expect(doc.status).toBe('draft');
+    expect(logSpy).toHaveBeenCalledWith(expect.stringContaining('successor'));
+  });
+
+  it('--export writes a labeled generated view and never touches authority', async () => {
+    const service = new PrdService({ projectPath: projectDir });
+    const created = await service.create({ name: 'Export Via CLI' });
+    const before = await service.get(created.id);
+
+    await runPrdy(['--export', created.id]);
+
+    const mdPath = path.join(projectDir, 'docs', 'prdy', `${created.id}.md`);
+    const md = await readFile(mdPath, 'utf8');
+    expect(md).toContain('generated view — do not edit');
+    expect(md).toContain(`prd-id: ${created.id}`);
+    expect(md).toContain('prd-version: 1');
+    expect(logSpy).toHaveBeenCalledWith(
+      expect.stringContaining(`Generated view written: ${mdPath}`),
+    );
+
+    // Authority unchanged by the export.
+    expect(await service.get(created.id)).toEqual(before);
+  });
+});

packages/mosaic/src/commands/prdy.ts

@@ -2,6 +2,10 @@ import type { Command } from 'commander';
 import { withAuth } from './with-auth.js';
 import { fetchProjects } from '../tui/gateway-api.js';
 
+/**
+ * `mosaic prdy` — thin adapter over PrdService (@mosaicstack/prdy).
+ * All reads/writes go through the service; there is no local writer path.
+ */
 export function registerPrdyCommand(program: Command) {
   const cmd = program
     .command('prdy')
@@ -9,12 +13,18 @@ export function registerPrdyCommand(program: Command) {
     .option('-g, --gateway <url>', 'Gateway URL', 'http://localhost:14242')
     .option('--init [name]', 'Create a new PRD')
     .option('--update [name]', 'Update an existing PRD')
+    .option('--import <file>', 'Import a YAML PRD document (validated, conflict-aware)')
+    .option('--accept-successor', 'With --import: accept a conflicted import as next version')
+    .option('--export [id]', 'Export a PRD as a labeled generated-view Markdown file')
     .option('--project <idOrName>', 'Scope to project')
     .action(
       async (opts: {
         gateway: string;
         init?: string | boolean;
         update?: string | boolean;
+        import?: string;
+        acceptSuccessor?: boolean;
+        export?: string | boolean;
         project?: string;
       }) => {
         // Detect project context when --project flag is provided
@@ -31,20 +41,69 @@ export function registerPrdyCommand(program: Command) {
           }
         }
 
+        const { PrdService, runPrdWizard } = await import('@mosaicstack/prdy');
+        const service = new PrdService({ projectPath: process.cwd() });
+
         try {
-          const { runPrdWizard } = await import('@mosaicstack/prdy');
+          if (opts.import !== undefined) {
+            const input = { filePath: opts.import };
+
+            if (opts.acceptSuccessor) {
+              const successor = await service.acceptSuccessor(input);
+              console.log(
+                `Import accepted as successor: ${successor.id} v${successor.version} (status: ${successor.status})`,
+              );
+              return;
+            }
+
+            const result = await service.importDocument(input);
+            console.log(
+              result.kind === 'created'
+                ? `Imported PRD ${result.document.id} v${result.document.version} (status: ${result.document.status})`
+                : `PRD ${result.document.id} already present with identical content — nothing to do.`,
+            );
+            return;
+          }
+
+          if (opts.export !== undefined) {
+            const id =
+              typeof opts.export === 'string' && opts.export.length > 0 ? opts.export : undefined;
+            const result = await service.exportMarkdown({ id });
+            console.log(
+              `Generated view written: ${result.filePath} (source authority: YAML under docs/prdy/ — do not edit the Markdown)`,
+            );
+            return;
+          }
+
           const name =
             typeof opts.init === 'string'
               ? opts.init
               : typeof opts.update === 'string'
                 ? opts.update
                 : 'untitled';
-          await runPrdWizard({
+
-            name,
+          if (process.stdout.isTTY) {
-            projectPath: process.cwd(),
+            await runPrdWizard({
-            interactive: true,
+              name,
-          });
+              projectPath: process.cwd(),
+              interactive: true,
+            });
+            return;
+          }
+
+          // Non-interactive fallback routes through the service directly.
+          const doc = await service.create({ name });
+          console.log(`PRD created: ${doc.id} v${doc.version} (status: ${doc.status})`);
         } catch (err) {
+          if (err instanceof Error && err.name === 'PrdImportConflictError') {
+            const conflict = err as { proposal?: { version?: number } };
+            console.error(`${err.message}`);
+            console.error(
+              `Original PRD left untouched. To accept the proposed successor (v${conflict.proposal?.version}), re-run with --accept-successor.`,
+            );
+            process.exit(1);
+          }
+
           console.error(`PRD wizard failed: ${err instanceof Error ? err.message : String(err)}`);
           process.exit(1);
         }

packages/prdy/src/cli.ts

@@ -1,6 +1,6 @@
 import { Command } from 'commander';
 
-import { createPrd, listPrds, loadPrd } from './prd.js';
+import { PrdService } from './service.js';
 import { runPrdWizard } from './wizard.js';
 
 interface InitCommandOptions {
@@ -18,6 +18,22 @@ interface ShowCommandOptions {
   readonly id?: string;
 }
 
+interface ImportCommandOptions {
+  readonly project: string;
+  readonly file: string;
+  readonly acceptSuccessor?: boolean;
+}
+
+interface ExportCommandOptions {
+  readonly project: string;
+  readonly id?: string;
+  readonly out?: string;
+}
+
+function serviceFor(project: string): PrdService {
+  return new PrdService({ projectPath: project });
+}
+
 export function buildPrdyCli(): Command {
   const program = new Command();
   program.name('mosaic').description('Mosaic CLI').exitOverride();
@@ -38,11 +54,9 @@ export function buildPrdyCli(): Command {
             template: options.template,
             interactive: true,
           })
-        : await createPrd({
+        : await serviceFor(options.project).create({
             name: options.name,
-            projectPath: options.project,
             template: options.template,
-            interactive: false,
           });
 
       console.log(
@@ -52,6 +66,7 @@ export function buildPrdyCli(): Command {
             id: doc.id,
             title: doc.title,
             status: doc.status,
+            version: doc.version,
             projectPath: doc.projectPath,
           },
           null,
@@ -65,7 +80,7 @@ export function buildPrdyCli(): Command {
     .description('List PRD documents for a project')
     .requiredOption('--project <path>', 'Project path')
     .action(async (options: ListCommandOptions) => {
-      const docs = await listPrds(options.project);
+      const docs = await serviceFor(options.project).list();
       console.log(JSON.stringify(docs, null, 2));
     });
 
@@ -75,20 +90,65 @@ export function buildPrdyCli(): Command {
     .requiredOption('--project <path>', 'Project path')
     .option('--id <id>', 'PRD document id')
     .action(async (options: ShowCommandOptions) => {
-      if (options.id !== undefined) {
+      const doc = await serviceFor(options.project).get(options.id);
-        const docs = await listPrds(options.project);
+      console.log(JSON.stringify(doc, null, 2));
-        const match = docs.find((doc) => doc.id === options.id);
+    });
 
-        if (match === undefined) {
+  prdy
-          throw new Error(`PRD id not found: ${options.id}`);
+    .command('import')
-        }
+    .description('Import a YAML PRD document (validated; conflicts propose a successor)')
+    .requiredOption('--project <path>', 'Project path')
+    .requiredOption('--file <file>', 'Path to YAML PRD document')
+    .option('--accept-successor', 'Accept a conflicted import as the next version')
+    .action(async (options: ImportCommandOptions) => {
+      const service = serviceFor(options.project);
+      const input = { filePath: options.file };
 
-        console.log(JSON.stringify(match, null, 2));
+      if (options.acceptSuccessor) {
+        const successor = await service.acceptSuccessor(input);
+        console.log(
+          JSON.stringify(
+            {
+              ok: true,
+              outcome: 'successor-accepted',
+              id: successor.id,
+              version: successor.version,
+            },
+            null,
+            2,
+          ),
+        );
         return;
       }
 
-      const doc = await loadPrd(options.project);
+      const result = await service.importDocument(input);
-      console.log(JSON.stringify(doc, null, 2));
+      console.log(
+        JSON.stringify(
+          {
+            ok: true,
+            outcome: result.kind,
+            id: result.document.id,
+            version: result.document.version,
+            status: result.document.status,
+          },
+          null,
+          2,
+        ),
+      );
+    });
+
+  prdy
+    .command('export')
+    .description('Render a PRD to a labeled generated-view Markdown file')
+    .requiredOption('--project <path>', 'Project path')
+    .option('--id <id>', 'PRD document id')
+    .option('--out <path>', 'Output path (default docs/prdy/<id>.md)')
+    .action(async (options: ExportCommandOptions) => {
+      const result = await serviceFor(options.project).exportMarkdown({
+        id: options.id,
+        outPath: options.out,
+      });
+      console.log(JSON.stringify({ ok: true, filePath: result.filePath }, null, 2));
     });
 
   return program;

packages/prdy/src/index.ts

@@ -1,12 +1,35 @@
-export { createPrd, loadPrd, savePrd, listPrds } from './prd.js';
+// PrdService is the single authority surface for PRD documents. The raw store
+// writers (createPrd/savePrd) are deliberately NOT exported: every mutation
+// goes through the service so there is no second writer path.
+export { loadPrd, listPrds, parsePrdDocument } from './prd.js';
 export { runPrdWizard } from './wizard.js';
 export { buildPrdyCli, runPrdyCli } from './cli.js';
 export { BUILTIN_PRD_TEMPLATES, resolveTemplate } from './templates.js';
+export {
+  PrdService,
+  PRD_GENERATED_VIEW_LABEL,
+  PrdError,
+  PrdNotFoundError,
+  PrdUpdateError,
+  PrdImportInvalidError,
+  PrdImportConflictError,
+} from './service.js';
 export type {
   PrdStatus,
   PrdTemplate,
   PrdTemplateSection,
   PrdSection,
+  PrdMissionLinkage,
   PrdDocument,
   CreatePrdOptions,
+  PrdServiceOptions,
+  PrdCreateInput,
+  PrdSectionPatch,
+  PrdUpdateInput,
+  PrdLinkMissionInput,
+  PrdPlanForMissionInput,
+  PrdExportInput,
+  PrdExportResult,
+  PrdImportInput,
+  PrdImportResult,
 } from './types.js';

packages/prdy/src/prd.ts

@@ -17,17 +17,49 @@ const prdSectionSchema = z.object({
   fields: z.record(z.string(), z.string()),
 });
 
+const prdMissionLinkageSchema = z.object({
+  missionId: z.string().min(1),
+  missionVersion: z.string().min(1),
+  prdVersion: z.number().int().min(1),
+  requirementIds: z.array(z.string()),
+  linkedAt: z.string().datetime(),
+});
+
 const prdDocumentSchema = z.object({
   id: z.string().min(1),
   title: z.string().min(1),
   status: z.enum(['draft', 'review', 'approved', 'archived']),
   projectPath: z.string().min(1),
   template: z.string().min(1),
+  // Defaults keep documents written by older prdy versions loadable.
+  version: z.number().int().min(1).default(1),
   sections: z.array(prdSectionSchema),
+  missions: z.array(prdMissionLinkageSchema).default([]),
   createdAt: z.string().datetime(),
   updatedAt: z.string().datetime(),
 });
 
+/** YAML timestamp scalars are parsed as Date by some emitters — normalize to ISO strings. */
+function coerceTimestamps(value: unknown): unknown {
+  if (value instanceof Date) {
+    return value.toISOString();
+  }
+  if (Array.isArray(value)) {
+    return value.map(coerceTimestamps);
+  }
+  if (typeof value === 'object' && value !== null) {
+    return Object.fromEntries(
+      Object.entries(value).map(([key, entry]) => [key, coerceTimestamps(entry)]),
+    );
+  }
+  return value;
+}
+
+/** Validate an unknown value as a PRD document (throws zod errors on failure). */
+export function parsePrdDocument(value: unknown): PrdDocument {
+  return prdDocumentSchema.parse(coerceTimestamps(value)) as PrdDocument;
+}
+
 function expandHome(projectPath: string): string {
   if (!projectPath.startsWith('~')) {
     return projectPath;
@@ -74,6 +106,8 @@ function prdDirectory(projectPath: string): string {
   return path.join(projectPath, PRD_DIRECTORY);
 }
 
+export { prdDirectory };
+
 function prdFilePath(projectPath: string, id: string): string {
   return path.join(prdDirectory(projectPath), `${id}.yaml`);
 }
@@ -113,11 +147,13 @@ export async function createPrd(options: CreatePrdOptions): Promise<PrdDocument>
     status: 'draft',
     projectPath: resolvedProjectPath,
     template: template.id,
+    version: 1,
     sections: template.sections.map((section) => ({
       id: section.id,
       title: section.title,
       fields: Object.fromEntries(section.fields.map((field) => [field, ''])),
     })),
+    missions: [],
     createdAt: now,
     updatedAt: now,
   };
@@ -190,7 +226,7 @@ export async function listPrds(projectPath: string): Promise<PrdDocument[]> {
       throw new Error(`Failed to parse PRD file ${filePath}: ${String(error)}`);
     }
 
-    const document = prdDocumentSchema.parse(parsed);
+    const document = parsePrdDocument(parsed);
     documents.push(document);
   }
 

packages/prdy/src/service.spec.ts

@@ -0,0 +1,433 @@
+import { existsSync } from 'node:fs';
+import { mkdtemp, readFile, readdir, writeFile } from 'node:fs/promises';
+import os from 'node:os';
+import path from 'node:path';
+
+import yaml from 'js-yaml';
+import { beforeEach, describe, expect, it } from 'vitest';
+
+import {
+  PRD_GENERATED_VIEW_LABEL,
+  PrdImportConflictError,
+  PrdImportInvalidError,
+  PrdNotFoundError,
+  PrdService,
+  PrdUpdateError,
+} from './index.js';
+import type { PrdDocument } from './index.js';
+
+// ── Helpers ──────────────────────────────────────────────────────────────────
+
+let projectDir: string;
+
+async function makeProject(): Promise<string> {
+  return mkdtemp(path.join(os.tmpdir(), 'prdy-service-'));
+}
+
+function service(): PrdService {
+  return new PrdService({ projectPath: projectDir });
+}
+
+function storeDir(): string {
+  return path.join(projectDir, 'docs', 'prdy');
+}
+
+/** Handcraft a full, schema-valid PRD document for import scenarios. */
+function importFixture(overrides: Partial<PrdDocument> = {}): PrdDocument {
+  return {
+    id: 'imported-prd-20260101-000000',
+    title: 'Imported PRD',
+    status: 'draft',
+    projectPath: '/tmp/elsewhere',
+    template: 'software',
+    version: 1,
+    sections: [
+      { id: 'introduction', title: 'Introduction', fields: { context: '', objective: '' } },
+      {
+        id: 'scope-non-goals',
+        title: 'Scope / Non-Goals',
+        fields: { inScope: '', outOfScope: '' },
+      },
+    ],
+    missions: [],
+    createdAt: '2026-01-01T00:00:00.000Z',
+    updatedAt: '2026-01-01T00:00:00.000Z',
+    ...overrides,
+  };
+}
+
+async function writeImportFile(doc: PrdDocument): Promise<string> {
+  const filePath = path.join(projectDir, `${doc.id}.import.yaml`);
+  await writeFile(filePath, yaml.dump(doc), 'utf8');
+  return filePath;
+}
+
+beforeEach(async () => {
+  projectDir = await makeProject();
+});
+
+// ── Single authority store (AC: prdy path and mission path resolve to the
+//    SAME store under docs/prdy/ with stable ids/versions) ────────────────────
+
+describe('PrdService single authority store', () => {
+  it('persists PRDs from the prdy path and the mission path into the same docs/prdy store', async () => {
+    const direct = await service().create({ name: 'Direct PRD' });
+    const viaMission = await service().planForMission({
+      name: 'Mission PRD',
+      missionId: 'mission-1',
+      missionVersion: '2026-01-01T00:00:00.000Z',
+    });
+
+    const files = await readdir(storeDir());
+    expect(files).toContain(`${direct.id}.yaml`);
+    expect(files).toContain(`${viaMission.id}.yaml`);
+
+    // A fresh service instance (new process equivalent) resolves both.
+    const all = await service().list();
+    expect(all.map((doc) => doc.id).sort()).toEqual([direct.id, viaMission.id].sort());
+
+    // Stable versions: creation is v1; linkage writes do not bump content version.
+    expect((await service().get(direct.id)).version).toBe(1);
+    expect((await service().get(viaMission.id)).version).toBe(1);
+  });
+
+  it('round-trips documents through the store with identity intact', async () => {
+    const created = await service().create({ name: 'Round Trip', template: 'feature' });
+    const fresh = await service().get(created.id);
+
+    expect(fresh).toEqual(created);
+    expect(fresh.id).toBe(created.id);
+    expect(fresh.template).toBe('feature');
+    expect(fresh.status).toBe('draft');
+  });
+
+  it('throws a typed error for unknown ids and empty stores', async () => {
+    await expect(service().get('nope')).rejects.toBeInstanceOf(PrdNotFoundError);
+    await expect(service().get()).rejects.toBeInstanceOf(PrdNotFoundError);
+  });
+});
+
+// ── Mission linkage persistence (AC: linkage survives restart via fresh
+//    service instances) ────────────────────────────────────────────────────────
+
+describe('PrdService mission linkage', () => {
+  it('persists linkage and reads it back from a fresh service instance', async () => {
+    const created = await service().planForMission({
+      name: 'Linked PRD',
+      missionId: 'mission-42',
+      missionVersion: '2026-02-03T04:05:06.000Z',
+      requirementIds: ['FR-1', 'FR-2'],
+    });
+
+    // Fresh instance — nothing in memory from the creating call.
+    const links = await service().listMissionLinks(created.id);
+    expect(links).toHaveLength(1);
+    expect(links[0]).toMatchObject({
+      missionId: 'mission-42',
+      missionVersion: '2026-02-03T04:05:06.000Z',
+      prdVersion: 1,
+      requirementIds: ['FR-1', 'FR-2'],
+    });
+
+    // Linkage is carried in the YAML authority file itself.
+    const raw = await readFile(path.join(storeDir(), `${created.id}.yaml`), 'utf8');
+    const persisted = yaml.load(raw) as PrdDocument;
+    expect(persisted.missions[0]?.missionId).toBe('mission-42');
+    expect(persisted.missions[0]?.requirementIds).toEqual(['FR-1', 'FR-2']);
+  });
+
+  it('refreshes an existing linkage entry in place instead of duplicating', async () => {
+    const created = await service().planForMission({
+      name: 'Relink PRD',
+      missionId: 'mission-7',
+      missionVersion: 'v1',
+    });
+
+    await service().update({
+      id: created.id,
+      sections: [{ id: 'introduction', fields: { objective: 'Ship it' } }],
+    });
+
+    const relinked = await service().linkMission({
+      prdId: created.id,
+      missionId: 'mission-7',
+      missionVersion: 'v2',
+      requirementIds: ['NFR-1'],
+    });
+
+    expect(relinked.missions).toHaveLength(1);
+    expect(relinked.missions[0]).toMatchObject({ missionVersion: 'v2', prdVersion: 2 });
+  });
+
+  it('does not bump the content version when writing linkage', async () => {
+    const created = await service().create({ name: 'Stable Version' });
+    const linked = await service().linkMission({
+      prdId: created.id,
+      missionId: 'm',
+      missionVersion: 'v1',
+    });
+    expect(linked.version).toBe(1);
+  });
+});
+
+// ── Update semantics ──────────────────────────────────────────────────────────
+
+describe('PrdService update', () => {
+  it('applies section patches and bumps the content version', async () => {
+    const created = await service().create({ name: 'Updatable' });
+    const updated = await service().update({
+      id: created.id,
+      sections: [{ id: 'introduction', fields: { context: 'Some context', objective: 'Goal' } }],
+    });
+
+    expect(updated.version).toBe(2);
+    expect(updated.sections[0]?.fields).toMatchObject({
+      context: 'Some context',
+      objective: 'Goal',
+    });
+    expect((await service().get(created.id)).version).toBe(2);
+  });
+
+  it('refuses unknown section ids with a typed error', async () => {
+    const created = await service().create({ name: 'Strict' });
+    await expect(
+      service().update({ id: created.id, sections: [{ id: 'nope', fields: {} }] }),
+    ).rejects.toBeInstanceOf(PrdUpdateError);
+  });
+});
+
+// ── Markdown export is a labeled generated view, never authority ──────────────
+
+describe('PrdService exportMarkdown', () => {
+  it('writes a generated view carrying the label and source identity', async () => {
+    const created = await service().create({ name: 'Exported PRD' });
+    const result = await service().exportMarkdown({ id: created.id });
+
+    expect(result.filePath).toBe(path.join(storeDir(), `${created.id}.md`));
+    expect(result.content).toContain(PRD_GENERATED_VIEW_LABEL);
+    expect(result.content).toContain(`prd-id: ${created.id}`);
+    expect(result.content).toContain('prd-version: 1');
+    expect(result.content).toContain(`source-of-truth: docs/prdy/${created.id}.yaml`);
+  });
+
+  it('reflects the current version after updates', async () => {
+    const created = await service().create({ name: 'Versioned Export' });
+    await service().update({
+      id: created.id,
+      sections: [{ id: 'introduction', fields: { objective: 'v2 goal' } }],
+    });
+    const result = await service().exportMarkdown({ id: created.id });
+    expect(result.content).toContain('prd-version: 2');
+  });
+
+  it('NEGATIVE CONTROL: mutating the exported Markdown cannot change the authority', async () => {
+    const created = await service().create({ name: 'Guarded PRD' });
+    const before = structuredClone(await service().get(created.id));
+
+    const result = await service().exportMarkdown({ id: created.id });
+    await writeFile(
+      result.filePath,
+      `<!-- ${PRD_GENERATED_VIEW_LABEL} -->\n# FAKE\nprd-id: fake-id\nprd-version: 99\n`,
+      'utf8',
+    );
+
+    const after = await service().get(created.id);
+    expect(after).toEqual(before);
+    expect(after.version).toBe(1);
+    expect(after.title).toBe(before.title);
+  });
+
+  it('never parses Markdown files that sit in the store directory', async () => {
+    const created = await service().create({ name: 'Decoy Guard' });
+
+    // A decoy .md file with invalid YAML must be invisible to the store.
+    await writeFile(path.join(storeDir(), 'decoy.md'), 'not: [valid: yaml', 'utf8');
+    // And a decoy .yaml-named Markdown body must not silently validate either.
+    await service().exportMarkdown({ id: created.id });
+
+    const listed = await service().list();
+    expect(listed.map((doc) => doc.id)).toEqual([created.id]);
+    await expect(service().get(created.id)).resolves.toBeTruthy();
+  });
+});
+
+// ── Import: validated, conflict-aware, never silently merging ─────────────────
+
+describe('PrdService importDocument', () => {
+  it('creates a valid import through the service, as draft — validity is not approval', async () => {
+    const filePath = await writeImportFile(importFixture({ status: 'approved' }));
+
+    const result = await service().importDocument({ filePath });
+
+    expect(result.kind).toBe('created');
+    expect(result.document.id).toBe('imported-prd-20260101-000000');
+    expect(result.document.status).toBe('draft'); // structural validity ≠ approval
+    expect(result.document.version).toBe(1);
+
+    const persisted = await service().get('imported-prd-20260101-000000');
+    expect(persisted.status).toBe('draft');
+
+    const files = await readdir(storeDir());
+    expect(files).toContain('imported-prd-20260101-000000.yaml');
+  });
+
+  it('reports identical content as a no-op without writing', async () => {
+    const created = await service().create({ name: 'Existing PRD' });
+    const before = await readFile(path.join(storeDir(), `${created.id}.yaml`), 'utf8');
+
+    const filePath = await writeImportFile(importFixture({ ...created }));
+    const result = await service().importDocument({ filePath });
+
+    expect(result.kind).toBe('identical');
+    const after = await readFile(path.join(storeDir(), `${created.id}.yaml`), 'utf8');
+    expect(after).toBe(before);
+  });
+
+  it('refuses a conflicting import with a typed error, a proposed successor, and untouched bytes', async () => {
+    const existing = await service().create({ name: 'Authority PRD' });
+    await service().linkMission({
+      prdId: existing.id,
+      missionId: 'mission-keep',
+      missionVersion: 'v1',
+      requirementIds: ['FR-0'],
+    });
+    const beforeBytes = await readFile(path.join(storeDir(), `${existing.id}.yaml`), 'utf8');
+
+    const divergent = importFixture({
+      ...existing,
+      title: 'Divergent Title',
+      sections: [
+        {
+          id: 'introduction',
+          title: 'Introduction',
+          fields: { context: 'changed', objective: '' },
+        },
+      ],
+    });
+    const filePath = await writeImportFile(divergent);
+
+    const attempt = service().importDocument({ filePath });
+    let caught: unknown;
+    try {
+      await attempt;
+    } catch (error) {
+      caught = error;
+    }
+    expect(caught).toBeInstanceOf(PrdImportConflictError);
+
+    const error = caught as PrdImportConflictError;
+    expect(error.code).toBe('PRD_IMPORT_CONFLICT');
+    expect(error.existing.id).toBe(existing.id);
+    expect(error.proposal.version).toBe(existing.version + 1); // successor proposal
+    expect(error.proposal.status).toBe('draft');
+
+    // Original authority content untouched on disk.
+    const afterBytes = await readFile(path.join(storeDir(), `${existing.id}.yaml`), 'utf8');
+    expect(afterBytes).toBe(beforeBytes);
+  });
+
+  it('acceptSuccessor persists the proposal explicitly, carrying linkages forward', async () => {
+    const existing = await service().create({ name: 'Successor Base' });
+    await service().linkMission({
+      prdId: existing.id,
+      missionId: 'mission-keep',
+      missionVersion: 'v1',
+    });
+
+    const divergent = importFixture({
+      ...existing,
+      title: 'Accepted Successor Title',
+    });
+    const filePath = await writeImportFile(divergent);
+
+    const successor = await service().acceptSuccessor({ filePath });
+    expect(successor.id).toBe(existing.id);
+    expect(successor.version).toBe(existing.version + 1);
+    expect(successor.title).toBe('Accepted Successor Title');
+    expect(successor.status).toBe('draft');
+    expect(successor.missions.map((m) => m.missionId)).toEqual(['mission-keep']);
+
+    // Persisted for a fresh reader.
+    const fresh = await service().get(existing.id);
+    expect(fresh.version).toBe(2);
+    expect(fresh.title).toBe('Accepted Successor Title');
+  });
+
+  it('refuses structurally-invalid imports with a typed error and creates nothing', async () => {
+    const cases: Array<{ name: string; body: string }> = [
+      { name: 'missing-title.yaml', body: yaml.dump({ id: 'x', status: 'draft' }) },
+      {
+        name: 'bad-status.yaml',
+        body: yaml.dump(importFixture({ status: 'not-a-status' as PrdDocument['status'] })),
+      },
+      {
+        name: 'bad-version.yaml',
+        body: yaml.dump(importFixture({ version: 0 })),
+      },
+      { name: 'not-yaml.yaml', body: '::: not yaml [\n  - {' },
+    ];
+
+    for (const fixture of cases) {
+      const filePath = path.join(projectDir, fixture.name);
+      await writeFile(filePath, fixture.body, 'utf8');
+
+      await expect(service().importDocument({ filePath })).rejects.toBeInstanceOf(
+        PrdImportInvalidError,
+      );
+    }
+
+    // Nothing was created: the authority store does not even exist yet.
+    await expect(readdir(storeDir())).rejects.toMatchObject({ code: 'ENOENT' });
+  });
+
+  it('acceptSuccessor refuses when there is no existing document to succeed', async () => {
+    const filePath = await writeImportFile(importFixture());
+    await expect(service().acceptSuccessor({ filePath })).rejects.toBeInstanceOf(PrdNotFoundError);
+  });
+});
+
+// ── No second writer: no code path reads exported Markdown back into authority ─
+
+describe('no-second-writer invariant (source-level)', () => {
+  // Resolve the package source dir whether vitest runs from the package root
+  // (turbo/pnpm test) or from the worktree root.
+  function resolveSrcDir(): string {
+    const candidates = [path.resolve('src'), path.resolve('packages/prdy/src')];
+    return candidates.find((dir) => existsSync(path.join(dir, 'service.ts'))) ?? candidates[0]!;
+  }
+
+  const srcDir = resolveSrcDir();
+  const sourceFiles = [
+    'cli.ts',
+    'index.ts',
+    'prd.ts',
+    'service.ts',
+    'templates.ts',
+    'types.ts',
+    'wizard.ts',
+  ];
+
+  it('no source file in @mosaicstack/prdy reads a .md file', async () => {
+    for (const file of sourceFiles) {
+      const text = await readFile(path.join(srcDir, file), 'utf8');
+      const readLines = text
+        .split('\n')
+        .map((line) => line.trim())
+        .filter((line) => /readFile|readFileSync|createReadStream/.test(line));
+
+      for (const line of readLines) {
+        expect(line.includes('.md'), `${file} reads a Markdown file: ${line}`).toBe(false);
+      }
+    }
+  });
+
+  it('the mosaic prdy/mission adapters never read a .md file', async () => {
+    const adapterDir = path.resolve(srcDir, '..', '..', 'mosaic', 'src', 'commands');
+    for (const file of ['prdy.ts', 'mission.ts']) {
+      const text = await readFile(path.join(adapterDir, file), 'utf8');
+      expect(text.includes("'.md'") || text.includes('.md`'), `${file} references a .md path`).toBe(
+        false,
+      );
+    }
+  });
+});

packages/prdy/src/service.ts

@@ -0,0 +1,379 @@
+import { promises as fs } from 'node:fs';
+import path from 'node:path';
+
+import yaml from 'js-yaml';
+
+import { createPrd, listPrds, parsePrdDocument, prdDirectory, savePrd } from './prd.js';
+import type {
+  PrdCreateInput,
+  PrdDocument,
+  PrdExportInput,
+  PrdExportResult,
+  PrdImportInput,
+  PrdImportResult,
+  PrdLinkMissionInput,
+  PrdMissionLinkage,
+  PrdPlanForMissionInput,
+  PrdServiceOptions,
+  PrdUpdateInput,
+} from './types.js';
+
+/**
+ * PrdService is the SINGLE authority surface for PRD documents.
+ *
+ * Every mutation path (CLI wizard, `mosaic mission --plan`, import) routes
+ * through this service; the YAML store under `docs/prdy/` is the authority and
+ * exported Markdown is a generated view that no code path reads back.
+ */
+
+// ── Typed errors ───────────────────────────────────────────────────────────────
+
+export class PrdError extends Error {
+  constructor(
+    message: string,
+    readonly code: string,
+  ) {
+    super(message);
+    this.name = 'PrdError';
+  }
+}
+
+export class PrdNotFoundError extends PrdError {
+  constructor(message: string) {
+    super(message, 'PRD_NOT_FOUND');
+    this.name = 'PrdNotFoundError';
+  }
+}
+
+export class PrdUpdateError extends PrdError {
+  constructor(message: string) {
+    super(message, 'PRD_UPDATE_INVALID');
+    this.name = 'PrdUpdateError';
+  }
+}
+
+/** Structural refusal: the import payload failed schema validation. Nothing is written. */
+export class PrdImportInvalidError extends PrdError {
+  constructor(
+    message: string,
+    readonly issues?: string,
+  ) {
+    super(message, 'PRD_IMPORT_INVALID');
+    this.name = 'PrdImportInvalidError';
+  }
+}
+
+/**
+ * Conflict refusal: an existing PRD shares the imported id but the content
+ * diverges. Carries a PROPOSED successor (existing version + 1) that is only
+ * persisted via an explicit {@link PrdService.acceptSuccessor} call — import
+ * never overwrites and never merges.
+ */
+export class PrdImportConflictError extends PrdError {
+  constructor(
+    message: string,
+    readonly existing: PrdDocument,
+    readonly proposal: PrdDocument,
+  ) {
+    super(message, 'PRD_IMPORT_CONFLICT');
+    this.name = 'PrdImportConflictError';
+  }
+}
+
+// ── Service ────────────────────────────────────────────────────────────────────
+
+/** The generated-view label carried by every Markdown export. */
+export const PRD_GENERATED_VIEW_LABEL = 'generated view — do not edit';
+
+export class PrdService {
+  private readonly projectPath: string;
+
+  constructor(options: PrdServiceOptions) {
+    this.projectPath = options.projectPath;
+  }
+
+  /** Create a new PRD (version 1, draft) in the authority store. */
+  async create(input: PrdCreateInput): Promise<PrdDocument> {
+    return createPrd({
+      name: input.name,
+      projectPath: this.projectPath,
+      template: input.template,
+      interactive: false,
+    });
+  }
+
+  /** Read a PRD by id, or the most recently updated one. */
+  async get(id?: string): Promise<PrdDocument> {
+    const documents = await listPrds(this.projectPath);
+
+    if (id === undefined) {
+      const latest = documents[0];
+      if (latest === undefined) {
+        throw new PrdNotFoundError(`No PRD documents found under docs/prdy/ for this project`);
+      }
+      return latest;
+    }
+
+    const match = documents.find((doc) => doc.id === id);
+    if (match === undefined) {
+      throw new PrdNotFoundError(`PRD id not found: ${id}`);
+    }
+    return match;
+  }
+
+  /** List all PRDs in the authority store (most recently updated first). */
+  async list(): Promise<PrdDocument[]> {
+    return listPrds(this.projectPath);
+  }
+
+  /**
+   * Apply section field patches and bump the content version.
+   * Linkage entries are preserved; linkage writes do NOT bump the version.
+   */
+  async update(input: PrdUpdateInput): Promise<PrdDocument> {
+    const doc = await this.get(input.id);
+
+    for (const patch of input.sections) {
+      const section = doc.sections.find((candidate) => candidate.id === patch.id);
+      if (section === undefined) {
+        throw new PrdUpdateError(`Unknown section id: ${patch.id}`);
+      }
+      for (const [field, value] of Object.entries(patch.fields)) {
+        if (!(field in section.fields)) {
+          throw new PrdUpdateError(`Unknown field "${field}" on section "${patch.id}"`);
+        }
+        section.fields[field] = value;
+      }
+    }
+
+    doc.version += 1;
+    doc.updatedAt = new Date().toISOString();
+    await savePrd(doc);
+    return doc;
+  }
+
+  /**
+   * Record (or refresh) a mission ↔ PRD linkage on the PRD document.
+   * Persisted in the YAML authority, so it survives restarts.
+   */
+  async linkMission(input: PrdLinkMissionInput): Promise<PrdDocument> {
+    const doc = await this.get(input.prdId);
+    return this.applyLinkage(doc, input);
+  }
+
+  /** Read back the mission linkages recorded on a PRD. */
+  async listMissionLinks(prdId?: string): Promise<PrdMissionLinkage[]> {
+    const doc = await this.get(prdId);
+    return doc.missions;
+  }
+
+  /**
+   * Mission planning path: create a PRD for a mission AND persist the
+   * mission↔PRD linkage in a single authority write.
+   */
+  async planForMission(input: PrdPlanForMissionInput): Promise<PrdDocument> {
+    const doc = await this.create({ name: input.name, template: input.template });
+    return this.applyLinkage(doc, {
+      prdId: doc.id,
+      missionId: input.missionId,
+      missionVersion: input.missionVersion,
+      requirementIds: input.requirementIds,
+    });
+  }
+
+  /**
+   * Render the PRD to a Markdown GENERATED VIEW.
+   *
+   * The output carries source identity (PRD id + version + generated-view
+   * label). It is written under `docs/prdy/<id>.md` and is NEVER read back:
+   * the authority store only loads `.yaml`/`.yml` files, and no code path in
+   * this package parses the exported Markdown.
+   */
+  async exportMarkdown(input?: PrdExportInput): Promise<PrdExportResult> {
+    const doc = await this.get(input?.id);
+    const content = renderMarkdown(doc);
+    const filePath = input?.outPath ?? path.join(prdDirectory(doc.projectPath), `${doc.id}.md`);
+
+    await fs.mkdir(path.dirname(filePath), { recursive: true });
+    await fs.writeFile(filePath, content, 'utf8');
+    return { filePath, content };
+  }
+
+  /**
+   * Import a YAML PRD document.
+   *
+   * Structural validation (zod) happens BEFORE anything is proposed or
+   * written. A structurally-valid import is persisted as `draft` — validity is
+   * NOT approval. If an existing PRD shares the id with divergent content, a
+   * typed {@link PrdImportConflictError} is thrown carrying a proposed
+   * successor; the original authority document is left byte-identical on disk.
+   */
+  async importDocument(input: PrdImportInput): Promise<PrdImportResult> {
+    const incoming = await this.readImportFile(input.filePath);
+
+    const existing = (await listPrds(this.projectPath)).find((doc) => doc.id === incoming.id);
+    if (existing === undefined) {
+      const document = this.buildImportedDocument(incoming);
+      await savePrd(document);
+      return { kind: 'created', document };
+    }
+
+    if (canonicalCore(existing) === canonicalCore(incoming)) {
+      return { kind: 'identical', document: existing };
+    }
+
+    throw new PrdImportConflictError(
+      `PRD id "${incoming.id}" already exists with divergent content — refusing to overwrite. ` +
+        `Proposed successor: version ${existing.version + 1} (draft). ` +
+        `Accept explicitly with acceptSuccessor().`,
+      existing,
+      this.buildSuccessor(existing, incoming),
+    );
+  }
+
+  /**
+   * Explicitly accept a conflicted import as a successor version of the
+   * existing PRD. Re-validates the source file before writing; the successor
+   * is persisted with status `draft` (acceptance of the import is not approval
+   * of the PRD) and the existing mission linkages are carried forward.
+   */
+  async acceptSuccessor(input: PrdImportInput): Promise<PrdDocument> {
+    const incoming = await this.readImportFile(input.filePath);
+
+    const existing = (await listPrds(this.projectPath)).find((doc) => doc.id === incoming.id);
+    if (existing === undefined) {
+      throw new PrdNotFoundError(
+        `No existing PRD with id "${incoming.id}" — use importDocument to create it`,
+      );
+    }
+
+    const successor = this.buildSuccessor(existing, incoming);
+    await savePrd(successor);
+    return successor;
+  }
+
+  // ── internals ──────────────────────────────────────────────────────────────
+
+  private async applyLinkage(doc: PrdDocument, input: PrdLinkMissionInput): Promise<PrdDocument> {
+    const entry: PrdMissionLinkage = {
+      missionId: input.missionId,
+      missionVersion: input.missionVersion,
+      prdVersion: doc.version,
+      requirementIds: input.requirementIds ?? [],
+      linkedAt: new Date().toISOString(),
+    };
+
+    // One entry per mission: refresh in place if the mission is already linked.
+    const index = doc.missions.findIndex((m) => m.missionId === entry.missionId);
+    if (index === -1) {
+      doc.missions.push(entry);
+    } else {
+      doc.missions[index] = entry;
+    }
+
+    // Linkage is mission-side metadata, not a content revision: bump the
+    // timestamp only so ids/versions stay stable for consumers.
+    doc.updatedAt = new Date().toISOString();
+    await savePrd(doc);
+    return doc;
+  }
+
+  private async readImportFile(filePath: string): Promise<PrdDocument> {
+    let raw: string;
+    try {
+      raw = await fs.readFile(filePath, 'utf8');
+    } catch (error) {
+      throw new PrdImportInvalidError(`Cannot read import file ${filePath}: ${String(error)}`);
+    }
+
+    let parsed: unknown;
+    try {
+      parsed = yaml.load(raw);
+    } catch (error) {
+      throw new PrdImportInvalidError(`Import file is not valid YAML: ${String(error)}`);
+    }
+
+    try {
+      return parsePrdDocument(parsed);
+    } catch (error) {
+      throw new PrdImportInvalidError(
+        `Import file failed PRD schema validation: ${filePath}`,
+        error instanceof Error ? error.message : String(error),
+      );
+    }
+  }
+
+  private buildImportedDocument(incoming: PrdDocument): PrdDocument {
+    const now = new Date().toISOString();
+    return {
+      ...incoming,
+      // The import lands in THIS project's authority store.
+      projectPath: this.projectPath,
+      // A structurally-valid import is not thereby approved.
+      status: 'draft',
+      version: 1,
+      missions: [],
+      createdAt: now,
+      updatedAt: now,
+    };
+  }
+
+  private buildSuccessor(existing: PrdDocument, incoming: PrdDocument): PrdDocument {
+    return {
+      ...incoming,
+      id: existing.id,
+      projectPath: existing.projectPath,
+      status: 'draft',
+      version: existing.version + 1,
+      missions: existing.missions,
+      createdAt: existing.createdAt,
+      updatedAt: new Date().toISOString(),
+    };
+  }
+}
+
+// ── Markdown rendering (generated view) ───────────────────────────────────────
+
+function canonicalCore(doc: PrdDocument): string {
+  return JSON.stringify([doc.title, doc.template, doc.sections]);
+}
+
+function renderMarkdown(doc: PrdDocument): string {
+  const lines: string[] = [
+    '<!--',
+    `${PRD_GENERATED_VIEW_LABEL}`,
+    `source-of-truth: docs/prdy/${doc.id}.yaml (YAML authority)`,
+    `prd-id: ${doc.id}`,
+    `prd-version: ${doc.version}`,
+    `generated-at: ${new Date().toISOString()}`,
+    '-->',
+    '',
+    `# ${doc.title}`,
+    '',
+    `**Status:** ${doc.status} · **Version:** ${doc.version} · **Template:** ${doc.template}`,
+    '',
+  ];
+
+  if (doc.missions.length > 0) {
+    lines.push('## Mission Linkage', '');
+    for (const mission of doc.missions) {
+      const requirements =
+        mission.requirementIds.length > 0 ? mission.requirementIds.join(', ') : 'none selected';
+      lines.push(
+        `- mission \`${mission.missionId}\` @ version \`${mission.missionVersion}\`` +
+          ` (linked at PRD v${mission.prdVersion}) — requirements: ${requirements}`,
+      );
+    }
+    lines.push('');
+  }
+
+  for (const section of doc.sections) {
+    lines.push(`## ${section.title}`, '');
+    for (const [field, value] of Object.entries(section.fields)) {
+      lines.push(`### ${field}`, '', value.trim().length > 0 ? value : '_Not set_.', '');
+    }
+  }
+
+  lines.push('---', '', `_End of generated view for ${doc.id} v${doc.version}._`, '');
+  return lines.join('\n');
+}

packages/prdy/src/types.ts

@@ -19,13 +19,31 @@ export interface PrdSection {
   fields: Record<string, string>;
 }
 
+/**
+ * Mission ↔ PRD linkage recorded on the PRD document (the YAML authority).
+ *
+ * `missionVersion` is the mission-side revision marker available to the CLI
+ * (the gateway exposes `updatedAt` for missions — there is no numeric mission
+ * version yet). `prdVersion` snapshots the PRD content version at link time.
+ */
+export interface PrdMissionLinkage {
+  missionId: string;
+  missionVersion: string;
+  prdVersion: number;
+  requirementIds: string[];
+  linkedAt: string;
+}
+
 export interface PrdDocument {
   id: string;
   title: string;
   status: PrdStatus;
   projectPath: string;
   template: string;
+  /** Content revision counter. Bumped by updates and accepted imports. */
+  version: number;
   sections: PrdSection[];
+  missions: PrdMissionLinkage[];
   createdAt: string;
   updatedAt: string;
 }
@@ -36,3 +54,60 @@ export interface CreatePrdOptions {
   template?: string;
   interactive?: boolean;
 }
+
+// ── PrdService surface (single authority entry point) ─────────────────────────
+
+export interface PrdServiceOptions {
+  projectPath: string;
+}
+
+export interface PrdCreateInput {
+  name: string;
+  template?: string;
+}
+
+export interface PrdSectionPatch {
+  id: string;
+  fields: Record<string, string>;
+}
+
+export interface PrdUpdateInput {
+  /** Defaults to the most recently updated PRD. */
+  id?: string;
+  sections: PrdSectionPatch[];
+}
+
+export interface PrdLinkMissionInput {
+  /** Defaults to the most recently updated PRD. */
+  prdId?: string;
+  missionId: string;
+  missionVersion: string;
+  requirementIds?: string[];
+}
+
+export interface PrdPlanForMissionInput extends PrdLinkMissionInput {
+  name: string;
+  template?: string;
+}
+
+export interface PrdExportInput {
+  /** Defaults to the most recently updated PRD. */
+  id?: string;
+  /** Override the generated-view output path. */
+  outPath?: string;
+}
+
+export interface PrdExportResult {
+  filePath: string;
+  content: string;
+}
+
+/** Discriminated result of a non-conflicting import. */
+export type PrdImportResult =
+  | { kind: 'created'; document: PrdDocument }
+  | { kind: 'identical'; document: PrdDocument };
+
+export interface PrdImportInput {
+  /** Path to a YAML-serialized PRD document (NOT the generated Markdown view). */
+  filePath: string;
+}

packages/prdy/src/wizard.ts

@@ -2,8 +2,8 @@ import path from 'node:path';
 
 import { cancel, intro, isCancel, outro, select, text } from '@clack/prompts';
 
-import { createPrd, savePrd } from './prd.js';
+import { PrdService } from './service.js';
-import type { CreatePrdOptions, PrdDocument } from './types.js';
+import type { CreatePrdOptions, PrdDocument, PrdSectionPatch } from './types.js';
 
 interface WizardAnswers {
   goals: string;
@@ -11,20 +11,41 @@ interface WizardAnswers {
   milestones: string;
 }
 
-function updateSectionField(doc: PrdDocument, sectionKeyword: string, value: string): void {
+/**
-  const section = doc.sections.find((candidate) => candidate.id.includes(sectionKeyword));
+ * Translate wizard answers into section patches using the same keyword
+ * matching the wizard always used (first section whose id contains the
+ * keyword, then first field whose name contains it, else first field).
+ */
+function buildWizardPatches(doc: PrdDocument, answers: WizardAnswers): PrdSectionPatch[] {
+  const bySection = new Map<string, PrdSectionPatch>();
 
-  if (section === undefined) {
+  const add = (keyword: string, value: string): void => {
-    return;
+    const section = doc.sections.find((candidate) => candidate.id.includes(keyword));
-  }
+    if (section === undefined) {
+      return;
+    }
 
-  const fieldName =
+    const fieldName =
-    Object.keys(section.fields).find((field) => field.toLowerCase().includes(sectionKeyword)) ??
+      Object.keys(section.fields).find((field) => field.toLowerCase().includes(keyword)) ??
-    Object.keys(section.fields)[0];
+      Object.keys(section.fields)[0];
 
-  if (fieldName !== undefined) {
+    if (fieldName === undefined || section.fields[fieldName] === value) {
-    section.fields[fieldName] = value;
+      return;
-  }
+    }
+
+    const existing = bySection.get(section.id);
+    if (existing === undefined) {
+      bySection.set(section.id, { id: section.id, fields: { [fieldName]: value } });
+    } else {
+      existing.fields[fieldName] = value;
+    }
+  };
+
+  add('goal', answers.goals);
+  add('constraint', answers.constraints);
+  add('milestone', answers.milestones);
+
+  return [...bySection.values()];
 }
 
 async function promptText(message: string, initialValue = ''): Promise<string> {
@@ -63,15 +84,10 @@ async function promptTemplate(template?: string): Promise<string> {
   return choice;
 }
 
-function applyWizardAnswers(doc: PrdDocument, answers: WizardAnswers): PrdDocument {
+/**
-  updateSectionField(doc, 'goal', answers.goals);
+ * Interactive PRD wizard. All writes go through PrdService — the wizard is a
-  updateSectionField(doc, 'constraint', answers.constraints);
+ * prompt layer, never a second writer path.
-  updateSectionField(doc, 'milestone', answers.milestones);
+ */
-
-  doc.updatedAt = new Date().toISOString();
-  return doc;
-}
-
 export async function runPrdWizard(options: CreatePrdOptions): Promise<PrdDocument> {
   intro('Mosaic PRD wizard');
 
@@ -82,20 +98,15 @@ export async function runPrdWizard(options: CreatePrdOptions): Promise<PrdDocume
   const constraints = await promptText('Key constraints');
   const milestones = await promptText('Planned milestones');
 
-  const doc = await createPrd({
+  const service = new PrdService({ projectPath: options.projectPath });
-    ...options,
+  const doc = await service.create({
     name,
     template,
-    interactive: true,
   });
 
-  const updated = applyWizardAnswers(doc, {
+  const patches = buildWizardPatches(doc, { goals, constraints, milestones });
-    goals,
+  const updated =
-    constraints,
+    patches.length > 0 ? await service.update({ id: doc.id, sections: patches }) : doc;
-    milestones,
-  });
-
-  await savePrd(updated);
 
   outro(`PRD created: ${path.join(updated.projectPath, 'docs', 'prdy', `${updated.id}.yaml`)}`);
 

packages/quality-rails/README.md

@@ -0,0 +1,63 @@
+# @mosaicstack/quality-rails
+
+Quality-rails scaffolding and — since RI-3-002 (RI-N4) — the **typed
+quality-rails evaluator**: the single authoritative producer of check verdicts
+for the checks it owns.
+
+## Evaluator (RI-N4)
+
+Every verdict is typed and fail-closed:
+
+```
+{ status: 'passed' | 'failed' | 'blocked' | 'error' | 'not-applicable',
+  checkId, checkVersion, subject, reason }
+```
+
+Missing implementations, missing inputs, unknown check ids, process errors,
+timeouts, and malformed probe output can never become `passed` or an
+unqualified skip — they surface as `blocked`/`error` with a reason (vocabulary
+mirrors MACP's `GateStatus` discipline).
+
+- Check definitions live as **data** with a version and a sha256 content
+  digest (`definitionDigest`); every recorded verdict names the definition
+  version that produced it.
+- Check sets are selected **per subject kind** (`node`, `python`, `rust`,
+  `monorepo`, `unknown`) via the versioned, digested check-set policy — this
+  repository (a `monorepo` subject) does not share the node template's file
+  list.
+- Shell probes stay **thin adapters**: the TS evaluator invokes them and owns
+  the verdict parsing (e.g. QC-20's planted-commit probe).
+
+### Owned checks
+
+| check id                    | canonical check | mechanism                                                              |
+| --------------------------- | --------------- | ---------------------------------------------------------------------- |
+| `qc-19-rails-files-present` | QC-19           | typed absorption of the former presence-only `check`/`doctor` loop     |
+| `qc-20-enforcement-verify`  | QC-20           | thin shell adapter (framework `verify.sh`); verdict parsing owned here |
+
+The canonical check ids QC-1..QC-21 and their dispositions are defined in
+`docs/release-integrity/probe-inventory.md` (the RI-3-001 inventory — the
+evaluator's input, not its output).
+
+### Usage
+
+```sh
+# CLI (same typed report as the programmatic API)
+node dist/cli.js quality-rails evaluate --project <path> [--probe-path <verify.sh>] [--json]
+node dist/cli.js quality-rails check --project <path> [--json]   # QC-19 only, fail-closed exit
+```
+
+```ts
+import { evaluateSubject } from '@mosaicstack/quality-rails';
+
+const report = await evaluateSubject({ subjectPath: '/path/to/project' });
+// report.state: 'passed' | 'failed' | 'blocked' | 'error'
+```
+
+`pnpm verify:release` invokes this evaluator as its `quality-rails` stage
+(canonical-only stage, QC-19 on the monorepo subject).
+
+## Scaffolding (pre-existing)
+
+`init` scaffolds rails files per detected kind/profile; `doctor` is advisory
+and reports typed states.

packages/quality-rails/src/cli-evaluate.spec.ts

@@ -0,0 +1,194 @@
+import { mkdir, mkdtemp, writeFile, chmod } from 'node:fs/promises';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+
+import { createQualityRailsCli } from './cli.js';
+import { QC_19_RAILS_FILES_PRESENT } from './evaluator/definitions.js';
+import { evaluateSubject } from './evaluator/runner.js';
+import type { EvaluationReport } from './evaluator/types.js';
+
+// CLI ↔ programmatic contract (RI-3-002): the same subject must produce the
+// same typed verdicts through every entry point the card adds — the
+// `evaluate`/`check` CLI surfaces and the `evaluateSubject` API.
+
+async function makeTempDir(): Promise<string> {
+  return mkdtemp(join(tmpdir(), 'qr-cli-'));
+}
+
+async function scaffoldNodeFixture(skip: string[] = []): Promise<string> {
+  const dir = await makeTempDir();
+  await writeFile(join(dir, 'package.json'), '{}\n', 'utf8');
+  for (const relativePath of [
+    '.eslintrc',
+    'biome.json',
+    '.githooks/pre-commit',
+    'PR-CHECKLIST.md',
+  ]) {
+    if (skip.includes(relativePath)) continue;
+    await mkdir(join(dir, relativePath, '..'), { recursive: true });
+    await writeFile(join(dir, relativePath), 'fixture\n', 'utf8');
+  }
+  return dir;
+}
+
+async function makePassingProbe(dir: string): Promise<string> {
+  const scriptPath = join(dir, 'probe-pass.sh');
+  await writeFile(
+    scriptPath,
+    [
+      '#!/bin/bash',
+      'echo "✅ PASS: Type errors blocked"',
+      'echo "✅ PASS: Lint errors blocked"',
+      'echo "Verification Summary"',
+      'exit 0',
+    ].join('\n') + '\n',
+    'utf8',
+  );
+  await chmod(scriptPath, 0o755);
+  return scriptPath;
+}
+
+describe('CLI entry points vs the programmatic evaluator', () => {
+  let logSpy: ReturnType<typeof vi.spyOn>;
+  let previousExitCode: string | number | undefined;
+
+  beforeEach(() => {
+    logSpy = vi.spyOn(console, 'log').mockImplementation(() => {});
+    previousExitCode = process.exitCode ?? undefined;
+  });
+
+  afterEach(() => {
+    logSpy.mockRestore();
+    process.exitCode = previousExitCode;
+  });
+
+  it('evaluate --json produces the SAME typed report as evaluateSubject (full check set + probe)', async () => {
+    const dir = await scaffoldNodeFixture();
+    const probePath = await makePassingProbe(dir);
+
+    const programmatic = await evaluateSubject({
+      subjectPath: dir,
+      inputs: { 'qc-20-enforcement-verify': { probePath } },
+    });
+
+    const program = createQualityRailsCli();
+    await program.parseAsync([
+      'node',
+      'cli.js',
+      'quality-rails',
+      'evaluate',
+      '--project',
+      dir,
+      '--probe-path',
+      probePath,
+      '--json',
+    ]);
+
+    const printed = logSpy.mock.calls.map((call) => String(call[0])).join('\n');
+    const cliReport = JSON.parse(printed) as EvaluationReport;
+    expect(cliReport).toEqual(programmatic);
+    expect(cliReport.state).toBe('passed');
+    expect(process.exitCode).toBe(0);
+  });
+
+  it('check --json produces the SAME QC-19 verdict as evaluateSubject (absorbed loop)', async () => {
+    const dir = await scaffoldNodeFixture(['biome.json', '.githooks/pre-commit']);
+
+    const programmatic = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_19_RAILS_FILES_PRESENT.id],
+    });
+    expect(programmatic.state).toBe('failed');
+
+    const program = createQualityRailsCli();
+    await program.parseAsync([
+      'node',
+      'cli.js',
+      'quality-rails',
+      'check',
+      '--project',
+      dir,
+      '--json',
+    ]);
+
+    const printed = logSpy.mock.calls.map((call) => String(call[0])).join('\n');
+    const cliReport = JSON.parse(printed) as EvaluationReport;
+    expect(cliReport).toEqual(programmatic);
+    expect(process.exitCode).toBe(1);
+  });
+
+  it('check on a complete subject exits 0 with a passed verdict', async () => {
+    const dir = await scaffoldNodeFixture();
+    const program = createQualityRailsCli();
+    await program.parseAsync([
+      'node',
+      'cli.js',
+      'quality-rails',
+      'check',
+      '--project',
+      dir,
+      '--json',
+    ]);
+    const printed = logSpy.mock.calls.map((call) => String(call[0])).join('\n');
+    const cliReport = JSON.parse(printed) as EvaluationReport;
+    expect(cliReport.state).toBe('passed');
+    expect(process.exitCode).toBe(0);
+  });
+
+  it('evaluate with an unknown check id exits 1 and reports error, never passed', async () => {
+    const dir = await scaffoldNodeFixture();
+    const program = createQualityRailsCli();
+    await program.parseAsync([
+      'node',
+      'cli.js',
+      'quality-rails',
+      'evaluate',
+      '--project',
+      dir,
+      '--check',
+      'qc-99-bogus',
+      '--json',
+    ]);
+    const printed = logSpy.mock.calls.map((call) => String(call[0])).join('\n');
+    const cliReport = JSON.parse(printed) as EvaluationReport;
+    expect(cliReport.results).toHaveLength(1);
+    const first = cliReport.results[0];
+    expect(first?.status).toBe('error');
+    expect(first?.reason).toContain('unknown check id');
+    expect(process.exitCode).toBe(1);
+  });
+
+  it('evaluate on a scaffold subject without --probe-path stays fail-closed (blocked, exit 1)', async () => {
+    const dir = await scaffoldNodeFixture();
+    const program = createQualityRailsCli();
+    await program.parseAsync([
+      'node',
+      'cli.js',
+      'quality-rails',
+      'evaluate',
+      '--project',
+      dir,
+      '--json',
+    ]);
+    const printed = logSpy.mock.calls.map((call) => String(call[0])).join('\n');
+    const cliReport = JSON.parse(printed) as EvaluationReport;
+    const qc20 = cliReport.results.find((r) => r.checkId === 'qc-20-enforcement-verify');
+    expect(qc20).toBeDefined();
+    expect(qc20?.status).toBe('blocked');
+    expect(qc20?.reason).toContain('probePath');
+    expect(cliReport.state).toBe('blocked');
+    expect(process.exitCode).toBe(1);
+  });
+
+  it('doctor stays advisory (no nonzero exit) but reports TYPED states, including blocked', async () => {
+    const dir = await scaffoldNodeFixture();
+    const program = createQualityRailsCli();
+    await program.parseAsync(['node', 'cli.js', 'quality-rails', 'doctor', '--project', dir]);
+
+    const printed = logSpy.mock.calls.map((call) => String(call[0])).join('\n');
+    expect(printed).toContain('blocked: qc-20-enforcement-verify');
+    expect(process.exitCode ?? 0).toBe(0);
+  });
+});

packages/quality-rails/src/cli.ts

@@ -1,5 +1,3 @@
-import { constants } from 'node:fs';
-import { access } from 'node:fs/promises';
 import { resolve } from 'node:path';
 import { fileURLToPath } from 'node:url';
 
@@ -8,18 +6,12 @@ import { Command } from 'commander';
 import { detectProjectKind } from './detect.js';
 import { scaffoldQualityRails } from './scaffolder.js';
 import type { ProjectKind, QualityProfile, RailsConfig } from './types.js';
+import { QC_19_RAILS_FILES_PRESENT } from './evaluator/definitions.js';
+import { evaluateSubject } from './evaluator/runner.js';
+import type { EvaluationReport } from './evaluator/types.js';
 
 const VALID_PROFILES: readonly QualityProfile[] = ['strict', 'standard', 'minimal'];
 
-async function fileExists(filePath: string): Promise<boolean> {
-  try {
-    await access(filePath, constants.F_OK);
-    return true;
-  } catch {
-    return false;
-  }
-}
-
 function parseProfile(rawProfile: string): QualityProfile {
   if (VALID_PROFILES.includes(rawProfile as QualityProfile)) {
     return rawProfile as QualityProfile;
@@ -59,22 +51,6 @@ function defaultFormatters(kind: ProjectKind): string[] {
   return [];
 }
 
-function expectedFilesForKind(kind: ProjectKind): string[] {
-  if (kind === 'node') {
-    return ['.eslintrc', 'biome.json', '.githooks/pre-commit', 'PR-CHECKLIST.md'];
-  }
-
-  if (kind === 'python') {
-    return ['pyproject.toml', '.githooks/pre-commit', 'PR-CHECKLIST.md'];
-  }
-
-  if (kind === 'rust') {
-    return ['rustfmt.toml', '.githooks/pre-commit', 'PR-CHECKLIST.md'];
-  }
-
-  return ['.githooks/pre-commit', 'PR-CHECKLIST.md'];
-}
-
 function printScaffoldResult(
   config: RailsConfig,
   filesWritten: string[],
@@ -106,6 +82,20 @@ function printScaffoldResult(
   }
 }
 
+function printEvaluationReport(report: EvaluationReport): void {
+  console.log(
+    `[quality-rails] evaluation for ${report.subject.path} (kind=${report.subject.kind}, check-set v${report.checkSetVersion})`,
+  );
+  for (const result of report.results) {
+    const reason = result.reason === undefined ? '' : ` — ${result.reason}`;
+    const digest = report.definitionDigests[result.checkId] ?? 'no digest';
+    console.log(
+      `  - ${result.status}: ${result.checkId} (v${result.checkVersion} [${digest}])${reason}`,
+    );
+  }
+  console.log(`[quality-rails] aggregate: ${report.state}`);
+}
+
 /**
  * Register quality-rails subcommands on an existing Commander program.
  * This avoids cross-package Commander version mismatches by using the
@@ -148,56 +138,93 @@ function buildQualityRailsCommand(qualityRails: Command): void {
       printScaffoldResult(config, result.filesWritten, result.warnings, result.commandsToRun);
     });
 
+  // `check` (QC-19) is ABSORBED by the RI-N4 evaluator: the presence loop
+  // that lived here is now the versioned, digested, typed check definition
+  // `qc-19-rails-files-present`. The CLI keeps its human surface (missing
+  // files listed, exit 1) and gains `--json` for the typed verdicts. Exit
+  // code is fail-closed: any non-green aggregate (failed/blocked/error) is 1.
   qualityRails
     .command('check')
     .requiredOption('--project <path>', 'Project path')
-    .action(async (options: { project: string }) => {
+    .option('--json', 'print the typed evaluation report as JSON')
+    .action(async (options: { project: string; json?: boolean }) => {
       const projectPath = resolve(options.project);
-      const kind = await detectProjectKind(projectPath);
+      const report = await evaluateSubject({
-      const expected = expectedFilesForKind(kind);
+        subjectPath: projectPath,
-      const missing: string[] = [];
+        checkIds: [QC_19_RAILS_FILES_PRESENT.id],
+      });
 
-      for (const relativePath of expected) {
+      if (options.json) {
-        const exists = await fileExists(resolve(projectPath, relativePath));
+        console.log(JSON.stringify(report));
-        if (!exists) {
+      } else {
-          missing.push(relativePath);
+        printEvaluationReport(report);
-        }
       }
 
-      if (missing.length > 0) {
+      process.exitCode = report.state === 'passed' ? 0 : 1;
-        console.error('[quality-rails] missing files:');
-        for (const relativePath of missing) {
-          console.error(`  - ${relativePath}`);
-        }
-        process.exitCode = 1;
-        return;
-      }
-
-      console.log(`[quality-rails] all expected files present for ${kind} project`);
     });
 
+  // `doctor` (QC-19) stays advisory (documented contract: a doctor that
+  // cannot fail), but now reports TYPED states — a blocked or failing rail is
+  // visible instead of silently printed as `ok`/`missing`.
   qualityRails
     .command('doctor')
     .requiredOption('--project <path>', 'Project path')
     .action(async (options: { project: string }) => {
       const projectPath = resolve(options.project);
-      const kind = await detectProjectKind(projectPath);
+      const report = await evaluateSubject({ subjectPath: projectPath });
-      const expected = expectedFilesForKind(kind);
 
       console.log(`[quality-rails] doctor for ${projectPath}`);
-      console.log(`detected project kind: ${kind}`);
+      console.log(`detected project kind: ${report.subject.kind}`);
-
+      for (const result of report.results) {
-      for (const relativePath of expected) {
+        const reason = result.reason === undefined ? '' : ` — ${result.reason}`;
-        const exists = await fileExists(resolve(projectPath, relativePath));
+        console.log(`  - ${result.status}: ${result.checkId}${reason}`);
-        console.log(`  - ${exists ? 'ok' : 'missing'}: ${relativePath}`);
       }
 
-      if (kind === 'unknown') {
+      if (report.subject.kind === 'unknown') {
         console.log(
           'recommendation: add package.json, pyproject.toml, or Cargo.toml for better defaults.',
         );
       }
     });
+
+  // `evaluate` is the canonical RI-N4 evaluator entry point: typed verdicts
+  // for the subject's full per-kind check set, same results as the
+  // programmatic API (evaluateSubject).
+  qualityRails
+    .command('evaluate')
+    .description('Run the typed quality-rails evaluator against a subject project')
+    .requiredOption('--project <path>', 'Project path')
+    .option('--check <id...>', 'restrict evaluation to these check ids')
+    .option(
+      '--probe-path <path>',
+      'path to the QC-20 behavioral probe script (framework verify.sh)',
+    )
+    .option('--json', 'print the typed evaluation report as JSON')
+    .action(
+      async (options: {
+        project: string;
+        check?: string[];
+        probePath?: string;
+        json?: boolean;
+      }) => {
+        const projectPath = resolve(options.project);
+        const report = await evaluateSubject({
+          subjectPath: projectPath,
+          checkIds: options.check,
+          inputs: options.probePath
+            ? { 'qc-20-enforcement-verify': { probePath: options.probePath } }
+            : undefined,
+        });
+
+        if (options.json) {
+          console.log(JSON.stringify(report));
+        } else {
+          printEvaluationReport(report);
+        }
+
+        process.exitCode = report.state === 'passed' ? 0 : 1;
+      },
+    );
 }
 
 export async function runQualityRailsCli(argv: string[] = process.argv): Promise<void> {

packages/quality-rails/src/evaluator/adapter.ts

@@ -0,0 +1,67 @@
+import { spawn } from 'node:child_process';
+import type { ChildProcess } from 'node:child_process';
+
+import type { AdapterOutcome, AdapterRequest, ProcessAdapter } from './types.js';
+
+/**
+ * Default thin process adapter (spawn-based). Runs a command to completion with
+ * a hard timeout and reports exit code + captured output — it owns NO verdict
+ * logic. Interpreting the outcome is always the check implementation's job.
+ */
+export function createSpawnProcessAdapter(): ProcessAdapter {
+  return {
+    run(request: AdapterRequest): Promise<AdapterOutcome> {
+      return new Promise((resolve) => {
+        let child: ChildProcess;
+        try {
+          child = spawn(request.file, request.args, {
+            cwd: request.cwd,
+            stdio: ['ignore', 'pipe', 'pipe'],
+          });
+        } catch (error) {
+          resolve({
+            ok: false,
+            kind: 'spawn-error',
+            message: error instanceof Error ? error.message : String(error),
+          });
+          return;
+        }
+
+        let stdout = '';
+        let stderr = '';
+        let settled = false;
+
+        const timer = setTimeout(() => {
+          if (settled) return;
+          settled = true;
+          child.kill('SIGKILL');
+          resolve({
+            ok: false,
+            kind: 'timeout',
+            message: `process timed out after ${request.timeoutMs}ms: ${request.file}`,
+          });
+        }, request.timeoutMs);
+
+        const settle = (outcome: AdapterOutcome): void => {
+          if (settled) return;
+          settled = true;
+          clearTimeout(timer);
+          resolve(outcome);
+        };
+
+        child.stdout?.on('data', (chunk: Buffer) => {
+          stdout += chunk.toString('utf8');
+        });
+        child.stderr?.on('data', (chunk: Buffer) => {
+          stderr += chunk.toString('utf8');
+        });
+        child.on('error', (error: Error) => {
+          settle({ ok: false, kind: 'spawn-error', message: error.message });
+        });
+        child.on('close', (code: number | null) => {
+          settle({ ok: true, exitCode: code, stdout, stderr });
+        });
+      });
+    },
+  };
+}

packages/quality-rails/src/evaluator/definitions.ts

@@ -0,0 +1,258 @@
+import { constants } from 'node:fs';
+import { access } from 'node:fs/promises';
+import { isAbsolute, resolve } from 'node:path';
+
+import { digestOfPolicy, digestOfSpec } from './digest.js';
+import type {
+  CheckContext,
+  CheckDefinition,
+  CheckDefinitionSpec,
+  CheckOutcome,
+  CheckSetPolicy,
+  CheckSetPolicySpec,
+  SubjectKind,
+} from './types.js';
+
+// Check definitions for the RI-N4 evaluator (card RI-3-002). Each definition is
+// DATA with a version and a content digest (see digest.ts); the executable
+// half is attached via defineCheck. Check-set SELECTION is per subject kind
+// (probe-inventory gap 7): this monorepo does not match the node template's
+// file list, so the QC-19 definition carries a distinct file set for the
+// `monorepo` subject kind and the policy selects checks per kind.
+
+export function defineCheck(
+  spec: CheckDefinitionSpec,
+  evaluate: (ctx: CheckContext) => Promise<CheckOutcome>,
+): CheckDefinition {
+  return { ...spec, definitionDigest: digestOfSpec(spec), evaluate };
+}
+
+async function fileExists(filePath: string): Promise<boolean> {
+  try {
+    await access(filePath, constants.F_OK);
+    return true;
+  } catch {
+    return false;
+  }
+}
+
+// ─── QC-19: downstream rails presence ────────────────────────────────────────
+//
+// Typed absorption of the former presence-only `quality-rails check` loop in
+// cli.ts. The scaffold-kind file lists below are carried over VERBATIM so the
+// evaluator's typed verdicts are parity-equivalent with the presence loop on
+// the same fixture; the `monorepo` list is new (per-subject check sets).
+
+const qc19Spec: CheckDefinitionSpec = {
+  id: 'qc-19-rails-files-present',
+  version: '1.0.0',
+  canonicalCheck: 'QC-19',
+  description:
+    'The subject still carries its quality-rails files. Typed absorption of the former presence-only check loop; presence is necessary, not sufficient (RI-N4).',
+  appliesTo: ['node', 'python', 'rust', 'monorepo', 'unknown'],
+  params: {
+    expectedFilesByKind: {
+      node: ['.eslintrc', 'biome.json', '.githooks/pre-commit', 'PR-CHECKLIST.md'],
+      python: ['pyproject.toml', '.githooks/pre-commit', 'PR-CHECKLIST.md'],
+      rust: ['rustfmt.toml', '.githooks/pre-commit', 'PR-CHECKLIST.md'],
+      monorepo: [
+        '.husky/pre-commit',
+        '.husky/pre-push',
+        'eslint.config.mjs',
+        '.prettierrc',
+        '.lintstagedrc',
+      ],
+      unknown: ['.githooks/pre-commit', 'PR-CHECKLIST.md'],
+    },
+  },
+};
+
+async function evaluateQc19(ctx: CheckContext): Promise<CheckOutcome> {
+  const byKind = ctx.params['expectedFilesByKind'] as Record<string, readonly string[]> | undefined;
+  if (byKind === undefined) {
+    return { status: 'error', reason: 'definition params missing expectedFilesByKind' };
+  }
+  const expected = byKind[ctx.subject.kind];
+  if (expected === undefined) {
+    // Fail-closed: an undefined file set for a declared subject kind is a
+    // definition gap, never a green outcome.
+    return {
+      status: 'blocked',
+      reason: `no expected-file set defined for subject kind '${ctx.subject.kind}'`,
+    };
+  }
+
+  const missing: string[] = [];
+  for (const relativePath of expected) {
+    if (!(await fileExists(resolve(ctx.subject.path, relativePath)))) {
+      missing.push(relativePath);
+    }
+  }
+  if (missing.length > 0) {
+    return {
+      status: 'failed',
+      reason: `missing rails files (${ctx.subject.kind}): ${missing.join(', ')}`,
+    };
+  }
+  return { status: 'passed' };
+}
+
+// ─── QC-20: downstream enforcement verification (behavioral probe) ──────────
+//
+// The planted-commit behavioral probe (framework tools/quality/scripts/verify.sh)
+// stays a THIN SHELL ADAPTER: the TS evaluator invokes it and OWNS the verdict
+// parsing (RI-N4: grep-on-output verdict logic moves into the typed evaluator).
+// Probe contract (verify.sh): exit 0 ⇔ every sub-probe passed, exit 1 ⇔ at
+// least one sub-probe failed; sub-probe verdicts appear as `PASS:` / `FAIL:`
+// marker lines and the script always prints a `Verification Summary` section.
+// Any deviation from that contract (other exit codes, unparseable output,
+// missing probe, process failure, timeout) is `error`/`blocked` — never
+// `passed`.
+
+const qc20Spec: CheckDefinitionSpec = {
+  id: 'qc-20-enforcement-verify',
+  version: '1.0.0',
+  canonicalCheck: 'QC-20',
+  description:
+    'The behavioral planted-commit probe runs against the subject and every sub-probe blocks as intended. The shell probe is a thin adapter; verdict parsing is owned by this evaluator.',
+  appliesTo: ['node', 'python', 'rust', 'unknown'],
+  params: {
+    command: 'bash',
+    timeoutMs: 120_000,
+    passMarker: 'PASS:',
+    failMarker: 'FAIL:',
+    summaryMarker: 'Verification Summary',
+  },
+};
+
+function linesWith(text: string, marker: string): string[] {
+  return text
+    .split('\n')
+    .map((line) => line.trim())
+    .filter((line) => line.includes(marker));
+}
+
+async function evaluateQc20(ctx: CheckContext): Promise<CheckOutcome> {
+  const rawProbePath = ctx.inputs['probePath'];
+  if (typeof rawProbePath !== 'string' || rawProbePath.trim().length === 0) {
+    return {
+      status: 'blocked',
+      reason:
+        'missing input: probePath — the behavioral probe script must be provided (e.g. the framework verify.sh)',
+    };
+  }
+  const probePath = isAbsolute(rawProbePath)
+    ? rawProbePath
+    : resolve(ctx.subject.path, rawProbePath);
+  if (!(await fileExists(probePath))) {
+    return { status: 'blocked', reason: `probe script not found: ${probePath}` };
+  }
+
+  const command = typeof ctx.params['command'] === 'string' ? ctx.params['command'] : 'bash';
+  const timeoutMs = typeof ctx.params['timeoutMs'] === 'number' ? ctx.params['timeoutMs'] : 120_000;
+  const passMarker =
+    typeof ctx.params['passMarker'] === 'string' ? ctx.params['passMarker'] : 'PASS:';
+  const failMarker =
+    typeof ctx.params['failMarker'] === 'string' ? ctx.params['failMarker'] : 'FAIL:';
+  const summaryMarker =
+    typeof ctx.params['summaryMarker'] === 'string'
+      ? ctx.params['summaryMarker']
+      : 'Verification Summary';
+
+  const outcome = await ctx.adapter.run({
+    file: command,
+    args: [probePath],
+    cwd: ctx.subject.path,
+    timeoutMs,
+  });
+
+  if (!outcome.ok) {
+    // Process error or timeout: the probe never produced a trustworthy result.
+    return {
+      status: 'error',
+      reason: `probe process ${outcome.kind}: ${outcome.message}`,
+    };
+  }
+
+  const output = `${outcome.stdout}\n${outcome.stderr}`;
+  const failLines = linesWith(output, failMarker);
+  const passLines = linesWith(output, passMarker);
+
+  if (outcome.exitCode === 0) {
+    // A green exit must be corroborated by a parseable green transcript:
+    // at least one pass marker, no fail markers, and the summary section.
+    if (passLines.length > 0 && failLines.length === 0 && output.includes(summaryMarker)) {
+      return { status: 'passed' };
+    }
+    return {
+      status: 'error',
+      reason: `malformed probe output: exit 0 without a parseable pass transcript (${passLines.length} pass markers, ${failLines.length} fail markers, summary ${output.includes(summaryMarker) ? 'present' : 'absent'})`,
+    };
+  }
+
+  if (outcome.exitCode === 1) {
+    if (failLines.length === 0) {
+      return {
+        status: 'error',
+        reason: 'malformed probe output: exit 1 without parseable FAIL markers',
+      };
+    }
+    return {
+      status: 'failed',
+      reason: `enforcement probe reported ${failLines.length} failing sub-probe(s): ${failLines.join(' | ')}`,
+    };
+  }
+
+  return {
+    status: 'error',
+    reason: `probe exited with unexpected code ${String(outcome.exitCode)} — outcome not interpretable`,
+  };
+}
+
+// ─── Per-subject check-set policy ───────────────────────────────────────────
+//
+// Gap 7 of the probe inventory: check sets must be selected per subject, not
+// one global list. Downstream scaffold kinds get the presence check plus the
+// behavioral probe (QC-20 blocks until a probePath input is provided — an
+// unverified subject can never evaluate green). The monorepo subject is this
+// repository itself: its rails are the husky hooks + shared lint/format
+// configs, covered by QC-19; the downstream planted-commit probe does not
+// apply to it (this repo's own commit gates are QC-13/QC-14, outside this
+// evaluator's owned checks).
+
+const checkSetPolicySpec: CheckSetPolicySpec = {
+  version: '1.0.0',
+  byKind: {
+    node: ['qc-19-rails-files-present', 'qc-20-enforcement-verify'],
+    python: ['qc-19-rails-files-present', 'qc-20-enforcement-verify'],
+    rust: ['qc-19-rails-files-present', 'qc-20-enforcement-verify'],
+    unknown: ['qc-19-rails-files-present', 'qc-20-enforcement-verify'],
+    monorepo: ['qc-19-rails-files-present'],
+  },
+};
+
+export const CHECK_SET_POLICY: CheckSetPolicy = {
+  ...checkSetPolicySpec,
+  policyDigest: digestOfPolicy(checkSetPolicySpec),
+};
+
+export const QC_19_RAILS_FILES_PRESENT = defineCheck(qc19Spec, evaluateQc19);
+export const QC_20_ENFORCEMENT_VERIFY = defineCheck(qc20Spec, evaluateQc20);
+
+/** Built-in check definitions, keyed by id. */
+export function builtInDefinitions(): CheckDefinition[] {
+  return [QC_19_RAILS_FILES_PRESENT, QC_20_ENFORCEMENT_VERIFY];
+}
+
+export function checkSetForKind(
+  kind: SubjectKind,
+  policy: CheckSetPolicy = CHECK_SET_POLICY,
+): readonly string[] {
+  const selected = policy.byKind[kind];
+  if (selected === undefined) {
+    // Fail-closed selection: an unknown kind yields an EMPTY set only to the
+    // caller; the runner treats an empty result list as `blocked`, never green.
+    return [];
+  }
+  return selected;
+}

packages/quality-rails/src/evaluator/digest.ts

@@ -0,0 +1,45 @@
+import { createHash } from 'node:crypto';
+
+import type { CheckDefinitionSpec, CheckSetPolicySpec } from './types.js';
+
+// Deterministic JSON: object keys sorted at every level so two specs with the
+// same content always produce the same bytes (and thus the same digest).
+export function canonicalJson(value: unknown): string {
+  if (value === null || typeof value !== 'object') {
+    return JSON.stringify(value);
+  }
+  if (Array.isArray(value)) {
+    return `[${value.map((entry) => canonicalJson(entry)).join(',')}]`;
+  }
+  const record = value as Record<string, unknown>;
+  const keys = Object.keys(record).sort();
+  return `{${keys.map((key) => `${JSON.stringify(key)}:${canonicalJson(record[key])}`).join(',')}}`;
+}
+
+/** sha256 over the canonical JSON of `value`. */
+export function digestContent(value: unknown): string {
+  return createHash('sha256').update(canonicalJson(value), 'utf8').digest('hex');
+}
+
+/**
+ * Content digest of a check definition: covers the declarative spec (id,
+ * version, canonical check, description, applicability, params) — everything a
+ * reviewer reasons about — while excluding the executable function object.
+ * Changing any covered field changes the digest, so a recorded digest always
+ * identifies exactly which definition content produced a verdict.
+ */
+export function digestOfSpec(spec: CheckDefinitionSpec): string {
+  return digestContent({
+    id: spec.id,
+    version: spec.version,
+    canonicalCheck: spec.canonicalCheck,
+    description: spec.description,
+    appliesTo: spec.appliesTo,
+    params: spec.params,
+  });
+}
+
+/** Content digest of the per-subject check-set policy. */
+export function digestOfPolicy(spec: CheckSetPolicySpec): string {
+  return digestContent(spec);
+}

packages/quality-rails/src/evaluator/evaluator.spec.ts

@@ -0,0 +1,525 @@
+import { mkdtemp, mkdir, writeFile, chmod } from 'node:fs/promises';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { fileURLToPath } from 'node:url';
+
+import { describe, expect, it } from 'vitest';
+
+import { createSpawnProcessAdapter } from './adapter.js';
+import {
+  builtInDefinitions,
+  CHECK_SET_POLICY,
+  checkSetForKind,
+  defineCheck,
+  QC_19_RAILS_FILES_PRESENT,
+  QC_20_ENFORCEMENT_VERIFY,
+} from './definitions.js';
+import { digestOfSpec } from './digest.js';
+import { aggregateState, evaluateSubject } from './runner.js';
+import type {
+  AdapterOutcome,
+  CheckDefinitionSpec,
+  CheckResult,
+  EvaluationReport,
+  ProcessAdapter,
+} from './types.js';
+
+// ─── helpers ────────────────────────────────────────────────────────────────
+
+function firstResult(report: EvaluationReport): CheckResult {
+  const result = report.results[0];
+  if (result === undefined) {
+    throw new Error('expected the report to contain at least one result');
+  }
+  return result;
+}
+
+async function makeTempDir(): Promise<string> {
+  return mkdtemp(join(tmpdir(), 'qr-evaluator-'));
+}
+
+async function writeProbeScript(dir: string, name: string, body: string): Promise<string> {
+  const scriptPath = join(dir, name);
+  await writeFile(scriptPath, `${body}\n`, 'utf8');
+  await chmod(scriptPath, 0o755);
+  return scriptPath;
+}
+
+/** Adapter stub that always returns the given outcome (no real process). */
+function stubAdapter(outcome: AdapterOutcome): ProcessAdapter {
+  return {
+    run: async () => outcome,
+  };
+}
+
+// VERBATIM copy of the pre-absorption presence loop (former cli.ts
+// expectedFilesForKind + fileExists loop). This is the PARITY ORACLE: the
+// evaluator's typed QC-19 verdict must agree with what the absorbed check
+// concluded on the same fixture.
+const LEGACY_EXPECTED: Record<'node' | 'python' | 'rust' | 'unknown', string[]> = {
+  node: ['.eslintrc', 'biome.json', '.githooks/pre-commit', 'PR-CHECKLIST.md'],
+  python: ['pyproject.toml', '.githooks/pre-commit', 'PR-CHECKLIST.md'],
+  rust: ['rustfmt.toml', '.githooks/pre-commit', 'PR-CHECKLIST.md'],
+  unknown: ['.githooks/pre-commit', 'PR-CHECKLIST.md'],
+};
+
+async function legacyPresenceLoop(projectPath: string, kind: keyof typeof LEGACY_EXPECTED) {
+  const missing: string[] = [];
+  for (const relativePath of LEGACY_EXPECTED[kind]) {
+    const fs = await import('node:fs/promises');
+    try {
+      await fs.access(join(projectPath, relativePath));
+    } catch {
+      missing.push(relativePath);
+    }
+  }
+  return missing;
+}
+
+async function scaffoldFixture(kind: keyof typeof LEGACY_EXPECTED, skip: string[] = []) {
+  const dir = await makeTempDir();
+  if (kind === 'node') {
+    await writeFile(join(dir, 'package.json'), '{}\n', 'utf8');
+  }
+  if (kind === 'python') {
+    await writeFile(join(dir, 'pyproject.toml'), '[project]\n', 'utf8');
+  }
+  if (kind === 'rust') {
+    await writeFile(join(dir, 'Cargo.toml'), '[package]\n', 'utf8');
+  }
+  for (const relativePath of LEGACY_EXPECTED[kind]) {
+    if (skip.includes(relativePath)) continue;
+    await mkdir(join(dir, relativePath, '..'), { recursive: true });
+    await writeFile(join(dir, relativePath), 'fixture\n', 'utf8');
+  }
+  return dir;
+}
+
+// ─── QC-19 parity: typed verdict == absorbed presence loop ──────────────────
+
+describe('QC-19 parity with the absorbed presence loop', () => {
+  const kinds: Array<keyof typeof LEGACY_EXPECTED> = ['node', 'python', 'rust', 'unknown'];
+
+  it.each(kinds)('positive fixture (%s): loop said ok ⇒ evaluator passed', async (kind) => {
+    const dir = await scaffoldFixture(kind);
+    const oracleMissing = await legacyPresenceLoop(dir, kind);
+    expect(oracleMissing).toEqual([]);
+
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_19_RAILS_FILES_PRESENT.id],
+    });
+    const result = report.results.find((r) => r.checkId === QC_19_RAILS_FILES_PRESENT.id);
+    expect(result?.status).toBe('passed');
+    expect(result?.reason).toBeUndefined();
+    expect(report.state).toBe('passed');
+  });
+
+  it.each(kinds)(
+    'negative fixture (%s): loop listed missing ⇒ evaluator failed with them',
+    async (kind) => {
+      const all = LEGACY_EXPECTED[kind];
+      const skip = all.slice(0, Math.max(1, all.length - 1)); // leave exactly 1 present
+      const dir = await scaffoldFixture(kind, skip);
+      const oracleMissing = await legacyPresenceLoop(dir, kind);
+      expect(oracleMissing.length).toBeGreaterThan(0);
+
+      const report = await evaluateSubject({
+        subjectPath: dir,
+        checkIds: [QC_19_RAILS_FILES_PRESENT.id],
+      });
+      const result = report.results.find((r) => r.checkId === QC_19_RAILS_FILES_PRESENT.id);
+      expect(result?.status).toBe('failed');
+      expect(report.state).toBe('failed');
+      for (const missingFile of oracleMissing) {
+        expect(result?.reason).toContain(missingFile);
+      }
+      // No false attribution: a present file must not be named in the reason.
+      const presentFile = all.find((file) => !skip.includes(file));
+      if (presentFile !== undefined) {
+        expect(result?.reason).not.toContain(` ${presentFile},`);
+      }
+    },
+  );
+});
+
+// ─── per-subject check sets (inventory gap 7) ────────────────────────────────
+
+describe('per-subject check sets', () => {
+  it('monorepo subject selects only QC-19 with the monorepo file set', async () => {
+    const dir = await makeTempDir();
+    await writeFile(join(dir, 'pnpm-workspace.yaml'), 'packages:\n  - packages/*\n', 'utf8');
+    for (const file of [
+      '.husky/pre-commit',
+      '.husky/pre-push',
+      'eslint.config.mjs',
+      '.prettierrc',
+      '.lintstagedrc',
+    ]) {
+      await mkdir(join(dir, file, '..'), { recursive: true });
+      await writeFile(join(dir, file), 'fixture\n', 'utf8');
+    }
+
+    const report = await evaluateSubject({ subjectPath: dir });
+    expect(report.subject.kind).toBe('monorepo');
+    expect(report.results.map((r) => r.checkId)).toEqual(['qc-19-rails-files-present']);
+    expect(report.state).toBe('passed');
+  });
+
+  it('a monorepo missing one of its rails files fails QC-19 (not the node list)', async () => {
+    const dir = await makeTempDir();
+    await writeFile(join(dir, 'pnpm-workspace.yaml'), 'packages:\n', 'utf8');
+    const report = await evaluateSubject({ subjectPath: dir });
+    const result = report.results.find((r) => r.checkId === QC_19_RAILS_FILES_PRESENT.id);
+    expect(result?.status).toBe('failed');
+    expect(result?.reason).toContain('.husky/pre-commit');
+    // The node-template list must NOT be applied to a monorepo subject.
+    expect(result?.reason).not.toContain('biome.json');
+  });
+
+  it('the policy selects the behavioral probe for scaffold kinds but not monorepo', () => {
+    expect(checkSetForKind('node')).toContain(QC_20_ENFORCEMENT_VERIFY.id);
+    expect(checkSetForKind('unknown')).toContain(QC_20_ENFORCEMENT_VERIFY.id);
+    expect(checkSetForKind('monorepo')).not.toContain(QC_20_ENFORCEMENT_VERIFY.id);
+    expect(CHECK_SET_POLICY.version).toBe('1.0.0');
+  });
+});
+
+// ─── negative controls (the point of the card) ───────────────────────────────
+
+describe('negative controls', () => {
+  it('unknown check id ⇒ error, never passed', async () => {
+    const dir = await scaffoldFixture('node');
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: ['qc-99-does-not-exist'],
+    });
+    expect(report.results).toHaveLength(1);
+    const result = firstResult(report);
+    expect(result.status).toBe('error');
+    expect(result.reason).toContain("unknown check id 'qc-99-does-not-exist'");
+    expect(result.status === 'passed').toBe(false);
+    expect(report.state).toBe('error');
+  });
+
+  it('missing subject (directory absent) ⇒ blocked for every check, never passed', async () => {
+    const report = await evaluateSubject({
+      subjectPath: join(tmpdir(), `qr-evaluator-absent-${Date.now()}`),
+    });
+    expect(report.results.length).toBeGreaterThan(0);
+    for (const result of report.results) {
+      expect(result.status).toBe('blocked');
+      expect(result.reason).toContain('subject directory does not exist');
+    }
+    expect(report.state).toBe('blocked');
+  });
+
+  it('QC-20 without probePath input ⇒ blocked, never passed', async () => {
+    const dir = await scaffoldFixture('node');
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_20_ENFORCEMENT_VERIFY.id],
+    });
+    const result = firstResult(report);
+    expect(result.status).toBe('blocked');
+    expect(result.reason).toContain('missing input: probePath');
+  });
+
+  it('QC-20 with a nonexistent probe script ⇒ blocked, never passed', async () => {
+    const dir = await scaffoldFixture('node');
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_20_ENFORCEMENT_VERIFY.id],
+      inputs: { 'qc-20-enforcement-verify': { probePath: join(dir, 'no-such-probe.sh') } },
+    });
+    expect(firstResult(report).status).toBe('blocked');
+    expect(firstResult(report).reason).toContain('probe script not found');
+  });
+
+  it('adapter process error (spawn failure) ⇒ error, never passed', async () => {
+    const dir = await scaffoldFixture('node');
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_20_ENFORCEMENT_VERIFY.id],
+      inputs: { 'qc-20-enforcement-verify': { probePath: join(dir, 'PR-CHECKLIST.md') } },
+      adapter: stubAdapter({ ok: false, kind: 'spawn-error', message: 'ENOENT bash' }),
+    });
+    expect(firstResult(report).status).toBe('error');
+    expect(firstResult(report).reason).toContain('probe process spawn-error');
+    expect(firstResult(report).status === 'passed').toBe(false);
+  });
+
+  it('adapter timeout ⇒ error, never passed', async () => {
+    const dir = await scaffoldFixture('node');
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_20_ENFORCEMENT_VERIFY.id],
+      inputs: { 'qc-20-enforcement-verify': { probePath: join(dir, 'PR-CHECKLIST.md') } },
+      adapter: stubAdapter({ ok: false, kind: 'timeout', message: 'timed out after 120000ms' }),
+    });
+    expect(firstResult(report).status).toBe('error');
+    expect(firstResult(report).reason).toContain('probe process timeout');
+  });
+
+  it('probe exit 1 with parseable FAIL markers ⇒ failed (interpretably red), never passed', async () => {
+    const dir = await scaffoldFixture('node');
+    const probe = await writeProbeScript(
+      dir,
+      'probe-fail.sh',
+      `echo "Test 1: ..."\necho "❌ FAIL: Type errors NOT blocked"\necho "Verification Summary"\nexit 1`,
+    );
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_20_ENFORCEMENT_VERIFY.id],
+      inputs: { 'qc-20-enforcement-verify': { probePath: probe } },
+      adapter: createSpawnProcessAdapter(),
+    });
+    expect(firstResult(report).status).toBe('failed');
+    expect(firstResult(report).reason).toContain('FAIL: Type errors NOT blocked');
+    expect(report.state).toBe('failed');
+  });
+
+  it('probe exit 1 WITHOUT parseable FAIL markers ⇒ malformed ⇒ error, never passed', async () => {
+    const dir = await scaffoldFixture('node');
+    const probe = await writeProbeScript(dir, 'probe-mute.sh', `echo "nothing to see"\nexit 1`);
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_20_ENFORCEMENT_VERIFY.id],
+      inputs: { 'qc-20-enforcement-verify': { probePath: probe } },
+      adapter: createSpawnProcessAdapter(),
+    });
+    expect(firstResult(report).status).toBe('error');
+    expect(firstResult(report).reason).toContain('malformed probe output');
+  });
+
+  it('probe exit 0 without a parseable pass transcript ⇒ malformed ⇒ error, never passed', async () => {
+    const dir = await scaffoldFixture('node');
+    const probe = await writeProbeScript(dir, 'probe-lie.sh', `echo "all good"\nexit 0`);
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_20_ENFORCEMENT_VERIFY.id],
+      inputs: { 'qc-20-enforcement-verify': { probePath: probe } },
+      adapter: createSpawnProcessAdapter(),
+    });
+    expect(firstResult(report).status).toBe('error');
+    expect(firstResult(report).reason).toContain('malformed probe output');
+    expect(firstResult(report).reason).toContain('exit 0');
+  });
+
+  it('probe exit 0 WITH fail markers ⇒ contradictory transcript ⇒ error, never passed', async () => {
+    const dir = await scaffoldFixture('node');
+    const probe = await writeProbeScript(
+      dir,
+      'probe-contradict.sh',
+      `echo "✅ PASS: one"\necho "❌ FAIL: two"\necho "Verification Summary"\nexit 0`,
+    );
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_20_ENFORCEMENT_VERIFY.id],
+      inputs: { 'qc-20-enforcement-verify': { probePath: probe } },
+      adapter: createSpawnProcessAdapter(),
+    });
+    expect(firstResult(report).status).toBe('error');
+  });
+
+  it('probe unexpected exit code (7) ⇒ error, never passed', async () => {
+    const dir = await scaffoldFixture('node');
+    const probe = await writeProbeScript(dir, 'probe-crash.sh', `echo "boom"\nexit 7`);
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_20_ENFORCEMENT_VERIFY.id],
+      inputs: { 'qc-20-enforcement-verify': { probePath: probe } },
+      adapter: createSpawnProcessAdapter(),
+    });
+    expect(firstResult(report).status).toBe('error');
+    expect(firstResult(report).reason).toContain('unexpected code 7');
+  });
+
+  it('check implementation throwing ⇒ error, never passed', async () => {
+    const dir = await scaffoldFixture('node');
+    const spec: CheckDefinitionSpec = {
+      id: 'test-throws',
+      version: '1.0.0',
+      canonicalCheck: 'QC-TEST',
+      description: 'sabotage-shaped definition that always throws',
+      appliesTo: ['node'],
+      params: {},
+    };
+    const throwing = defineCheck(spec, async () => {
+      throw new Error('kaboom');
+    });
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: ['test-throws'],
+      definitions: [throwing],
+    });
+    expect(firstResult(report).status).toBe('error');
+    expect(firstResult(report).reason).toContain('kaboom');
+  });
+
+  it('non-passed verdict without a reason ⇒ upgraded to error, never an unqualified skip', async () => {
+    const dir = await scaffoldFixture('node');
+    const spec: CheckDefinitionSpec = {
+      id: 'test-silent-fail',
+      version: '1.0.0',
+      canonicalCheck: 'QC-TEST',
+      description: 'returns failed without a reason',
+      appliesTo: ['node'],
+      params: {},
+    };
+    const silent = defineCheck(spec, async () => ({ status: 'failed' }));
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: ['test-silent-fail'],
+      definitions: [silent],
+    });
+    expect(firstResult(report).status).toBe('error');
+    expect(firstResult(report).reason).toContain('without a reason');
+  });
+
+  it('empty result list aggregates to blocked, never passed', () => {
+    expect(aggregateState([])).toBe('blocked');
+  });
+});
+
+// ─── QC-20 parity: typed verdict == shell probe's own conclusion ────────────
+
+describe('QC-20 parity with the shell probe contract', () => {
+  it('green transcript (exit 0) ⇒ evaluator passed', async () => {
+    const dir = await scaffoldFixture('node');
+    const probe = await writeProbeScript(
+      dir,
+      'probe-pass.sh',
+      [
+        'echo "✅ PASS: Type errors blocked"',
+        'echo "✅ PASS: any types blocked"',
+        'echo "✅ PASS: Lint errors blocked"',
+        'echo "Verification Summary"',
+        'echo "✅ Passed: 3"',
+        'exit 0',
+      ].join('\n'),
+    );
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_20_ENFORCEMENT_VERIFY.id],
+      inputs: { 'qc-20-enforcement-verify': { probePath: probe } },
+      adapter: createSpawnProcessAdapter(),
+    });
+    expect(firstResult(report).status).toBe('passed');
+    expect(report.state).toBe('passed');
+  });
+
+  it('the REAL framework verify.sh on a non-git subject concludes failed (exit 1) ⇒ evaluator failed', async () => {
+    // Real-probe parity: verify.sh without a git repo cannot block planted
+    // commits, exits 1 with FAIL markers — the evaluator must record exactly
+    // `failed` with those markers, matching the probe's own conclusion.
+    const realProbe = fileURLToPath(
+      new URL('../../../mosaic/framework/tools/quality/scripts/verify.sh', import.meta.url),
+    );
+    const dir = await makeTempDir(); // not a git repository, no hooks
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_20_ENFORCEMENT_VERIFY.id],
+      inputs: { 'qc-20-enforcement-verify': { probePath: realProbe } },
+      adapter: createSpawnProcessAdapter(),
+    });
+    const result = firstResult(report);
+    expect(result.status).toBe('failed');
+    expect(result.reason).toMatch(/FAIL:/);
+    expect(report.state).toBe('failed');
+  });
+});
+
+// ─── version / digest discipline ─────────────────────────────────────────────
+
+describe('versioned, digested check definitions', () => {
+  it('every verdict records the definition version that produced it', async () => {
+    const dir = await scaffoldFixture('node');
+    const report = await evaluateSubject({ subjectPath: dir });
+    for (const result of report.results) {
+      expect(result.checkVersion).toBe('1.0.0');
+    }
+    expect(report.checkSetVersion).toBe(CHECK_SET_POLICY.version);
+  });
+
+  it('the report records each definition’s content digest', async () => {
+    const dir = await scaffoldFixture('node');
+    const report = await evaluateSubject({ subjectPath: dir });
+    expect(report.definitionDigests[QC_19_RAILS_FILES_PRESENT.id]).toBe(
+      QC_19_RAILS_FILES_PRESENT.definitionDigest,
+    );
+    expect(report.definitionDigests[QC_20_ENFORCEMENT_VERIFY.id]).toBe(
+      QC_20_ENFORCEMENT_VERIFY.definitionDigest,
+    );
+  });
+
+  it('digests are stable for identical content', () => {
+    const spec = QC_19_RAILS_FILES_PRESENT;
+    expect(digestOfSpec(spec)).toBe(digestOfSpec(spec));
+  });
+
+  it('changing a definition’s content changes its digest', () => {
+    const base = { ...QC_19_RAILS_FILES_PRESENT } as CheckDefinitionSpec;
+    const baseDigest = digestOfSpec(base);
+
+    const changedParams: CheckDefinitionSpec = {
+      ...base,
+      params: {
+        expectedFilesByKind: {
+          ...(base.params['expectedFilesByKind'] as Record<string, string[]>),
+          node: ['.eslintrc', 'biome.json', '.githooks/pre-commit', 'PR-CHECKLIST.md', 'NEW.md'],
+        },
+      },
+    };
+    expect(digestOfSpec(changedParams)).not.toBe(baseDigest);
+
+    const changedVersion: CheckDefinitionSpec = { ...base, version: '1.1.0' };
+    expect(digestOfSpec(changedVersion)).not.toBe(baseDigest);
+  });
+
+  it('a definition with changed content produces a different recorded digest and version', async () => {
+    const dir = await scaffoldFixture('node');
+    const modified = defineCheck(
+      { ...QC_19_RAILS_FILES_PRESENT, version: '2.0.0' } as unknown as CheckDefinitionSpec,
+      async () => ({ status: 'passed' }),
+    );
+    const report = await evaluateSubject({
+      subjectPath: dir,
+      checkIds: [QC_19_RAILS_FILES_PRESENT.id],
+      definitions: [
+        modified,
+        ...builtInDefinitions().filter((d) => d.id !== QC_19_RAILS_FILES_PRESENT.id),
+      ],
+    });
+    const result = firstResult(report);
+    expect(result.checkVersion).toBe('2.0.0');
+    expect(report.definitionDigests[QC_19_RAILS_FILES_PRESENT.id]).toBe(modified.definitionDigest);
+    expect(modified.definitionDigest).not.toBe(QC_19_RAILS_FILES_PRESENT.definitionDigest);
+  });
+});
+
+// ─── aggregate state ordering (MACP-style discipline) ───────────────────────
+
+describe('aggregate state precedence', () => {
+  const result = (status: 'passed' | 'failed' | 'blocked' | 'error') => ({
+    status,
+    checkId: 'x',
+    checkVersion: '1.0.0',
+    subject: '/tmp/x',
+  });
+
+  it('all passed (with not-applicable) ⇒ passed', () => {
+    expect(
+      aggregateState([
+        result('passed'),
+        { ...result('passed'), status: 'not-applicable' as const },
+      ]),
+    ).toBe('passed');
+  });
+
+  it('error outranks blocked and failed; blocked outranks failed', () => {
+    expect(aggregateState([result('blocked'), result('error')])).toBe('error');
+    expect(aggregateState([result('failed'), result('blocked')])).toBe('blocked');
+    expect(aggregateState([result('passed'), result('failed')])).toBe('failed');
+  });
+});

packages/quality-rails/src/evaluator/runner.ts

@@ -0,0 +1,183 @@
+import { constants } from 'node:fs';
+import { access, stat } from 'node:fs/promises';
+import { join, resolve } from 'node:path';
+
+import { createSpawnProcessAdapter } from './adapter.js';
+import { builtInDefinitions, CHECK_SET_POLICY, checkSetForKind } from './definitions.js';
+import type {
+  AggregateState,
+  CheckResult,
+  CheckStatus,
+  EvaluateOptions,
+  EvaluationReport,
+  ProcessAdapter,
+  Subject,
+  SubjectKind,
+} from './types.js';
+import { detectProjectKind } from '../detect.js';
+
+async function pathExists(targetPath: string): Promise<boolean> {
+  try {
+    await access(targetPath, constants.F_OK);
+    return true;
+  } catch {
+    return false;
+  }
+}
+
+async function isDirectory(targetPath: string): Promise<boolean> {
+  try {
+    return (await stat(targetPath)).isDirectory();
+  } catch {
+    return false;
+  }
+}
+
+/**
+ * Subject-kind detection for the evaluator. Extends the scaffold detection
+ * (detect.ts) with the `monorepo` kind: a pnpm workspace is this repository's
+ * own subject shape and carries a different rails file set (probe-inventory
+ * gap 7 — check sets are per subject, not one global file list).
+ */
+export async function detectSubjectKind(subjectPath: string): Promise<SubjectKind> {
+  if (await pathExists(join(subjectPath, 'pnpm-workspace.yaml'))) {
+    return 'monorepo';
+  }
+  const kind = await detectProjectKind(subjectPath);
+  return kind;
+}
+
+/**
+ * Aggregate state, MACP-style discipline: `passed` only when at least one
+ * check produced a verdict AND every verdict is `passed` or an explicitly
+ * qualified `not-applicable`. Precedence is fail-closed: error > blocked >
+ * failed > passed; an empty result list aggregates to `blocked`.
+ */
+export function aggregateState(results: readonly CheckResult[]): AggregateState {
+  if (results.length === 0) {
+    return 'blocked';
+  }
+  const has = (status: CheckStatus): boolean => results.some((result) => result.status === status);
+  if (has('error')) {
+    return 'error';
+  }
+  if (has('blocked')) {
+    return 'blocked';
+  }
+  if (has('failed')) {
+    return 'failed';
+  }
+  return 'passed';
+}
+
+function reasonFrom(error: unknown): string {
+  return error instanceof Error ? error.message : String(error);
+}
+
+/**
+ * Evaluate one subject against a set of checks, producing typed verdicts.
+ *
+ * Fail-closed invariants (RI-N4):
+ * - unknown check id            → `error` (never passed)
+ * - subject directory absent    → every verdict `blocked`
+ * - check implementation threw  → `error`
+ * - non-passed without a reason → `error` (no unqualified skips)
+ * - check not applicable        → `not-applicable` WITH a reason
+ */
+export async function evaluateSubject(options: EvaluateOptions): Promise<EvaluationReport> {
+  const subjectPath = resolve(options.subjectPath);
+  const subject: Subject = {
+    path: subjectPath,
+    kind: await detectSubjectKind(subjectPath),
+  };
+
+  const definitions = options.definitions ?? builtInDefinitions();
+  const byId = new Map(definitions.map((definition) => [definition.id, definition]));
+  const requested = options.checkIds ?? checkSetForKind(subject.kind);
+  const adapter: ProcessAdapter = options.adapter ?? createSpawnProcessAdapter();
+
+  const results: CheckResult[] = [];
+  const definitionDigests: Record<string, string> = {};
+
+  for (const checkId of requested) {
+    const definition = byId.get(checkId);
+    if (definition === undefined) {
+      const known = definitions.map((entry) => entry.id).join(', ');
+      results.push({
+        status: 'error',
+        checkId,
+        checkVersion: 'unknown',
+        subject: subjectPath,
+        reason: `unknown check id '${checkId}' — no registered definition (known: ${known})`,
+      });
+      continue;
+    }
+
+    definitionDigests[checkId] = definition.definitionDigest;
+
+    if (!(await isDirectory(subjectPath))) {
+      results.push({
+        status: 'blocked',
+        checkId,
+        checkVersion: definition.version,
+        subject: subjectPath,
+        reason: `subject directory does not exist: ${subjectPath}`,
+      });
+      continue;
+    }
+
+    if (!definition.appliesTo.includes(subject.kind)) {
+      results.push({
+        status: 'not-applicable',
+        checkId,
+        checkVersion: definition.version,
+        subject: subjectPath,
+        reason: `check '${checkId}' does not apply to subject kind '${subject.kind}'`,
+      });
+      continue;
+    }
+
+    try {
+      const inputs = options.inputs?.[checkId] ?? {};
+      const outcome = await definition.evaluate({
+        subject,
+        params: definition.params,
+        inputs,
+        adapter,
+      });
+      if (outcome.status !== 'passed' && (outcome.reason === undefined || outcome.reason === '')) {
+        results.push({
+          status: 'error',
+          checkId,
+          checkVersion: definition.version,
+          subject: subjectPath,
+          reason: `check returned status '${outcome.status}' without a reason — treated as error`,
+        });
+        continue;
+      }
+      results.push({
+        status: outcome.status,
+        checkId,
+        checkVersion: definition.version,
+        subject: subjectPath,
+        reason: outcome.reason,
+      });
+    } catch (error) {
+      results.push({
+        status: 'error',
+        checkId,
+        checkVersion: definition.version,
+        subject: subjectPath,
+        reason: `check implementation threw: ${reasonFrom(error)}`,
+      });
+    }
+  }
+
+  return {
+    subject,
+    results,
+    definitionDigests,
+    checkSetVersion: CHECK_SET_POLICY.version,
+    state: aggregateState(results),
+  };
+}

packages/quality-rails/src/evaluator/types.ts

@@ -0,0 +1,148 @@
+// Evaluator core types — RI-N4 (card RI-3-002, SDLC-D-037 second half).
+//
+// The quality-rails evaluator is the SOLE authoritative producer of check
+// verdicts for the checks it owns. Every verdict is typed and fail-closed:
+// missing implementations, missing inputs, unknown check ids, process errors,
+// timeouts, and malformed adapter output can never become `passed` or an
+// unqualified skip — they surface as `blocked` or `error` with a reason.
+// (Vocabulary mirrors MACP's GateStatus discipline from packages/macp.)
+
+/**
+ * Typed verdict for a single check execution.
+ *
+ * - `passed` — the check really ran and its condition held.
+ * - `failed` — the check really ran and its condition did NOT hold.
+ * - `blocked` — the check could not run at all (missing subject, missing
+ *   input). Never a green outcome.
+ * - `error` — the check attempted to run but its outcome cannot be trusted
+ *   (unknown check id, implementation threw, process error, timeout, malformed
+ *   adapter output). Never a green outcome.
+ * - `not-applicable` — the check definition explicitly declares it does not
+ *   apply to this subject (a qualified skip, always with a reason).
+ */
+export type CheckStatus = 'passed' | 'failed' | 'blocked' | 'error' | 'not-applicable';
+
+/** Aggregate outcome, MACP-style: `passed` only when every result is green. */
+export type AggregateState = 'passed' | 'failed' | 'blocked' | 'error';
+
+/** Kinds of subjects the evaluator can assess. */
+export type SubjectKind = 'node' | 'python' | 'rust' | 'monorepo' | 'unknown';
+
+/**
+ * A single check verdict. This is the canonical result shape: `status`,
+ * `checkId`, `checkVersion`, `subject`, `reason`. `reason` is REQUIRED
+ * (enforced by the runner) for every status other than `passed`.
+ */
+export interface CheckResult {
+  status: CheckStatus;
+  checkId: string;
+  checkVersion: string;
+  subject: string;
+  reason?: string;
+}
+
+/** The project being evaluated. */
+export interface Subject {
+  /** Absolute path. */
+  path: string;
+  kind: SubjectKind;
+}
+
+/**
+ * The data half of a check definition. Definitions live as DATA with a version
+ * and a content digest (see `digestOfSpec`); the executable half is attached
+ * separately so the digest covers only reviewable, declarative content.
+ */
+export interface CheckDefinitionSpec {
+  /** Stable id, e.g. `qc-19-rails-files-present`. */
+  id: string;
+  /** Semver of this definition's data+semantics. */
+  version: string;
+  /** Canonical check id from docs/release-integrity/probe-inventory.md (QC-n). */
+  canonicalCheck: string;
+  description: string;
+  /** Subject kinds this check can assess (others yield `not-applicable`). */
+  appliesTo: readonly SubjectKind[];
+  /** Declarative parameters (file lists, markers, timeouts) — digest-covered. */
+  params: Record<string, unknown>;
+}
+
+/** A fully assembled check definition: spec + digest + implementation. */
+export interface CheckDefinition extends CheckDefinitionSpec {
+  /** sha256 content digest of the spec (canonical JSON projection). */
+  definitionDigest: string;
+  evaluate(ctx: CheckContext): Promise<CheckOutcome>;
+}
+
+/** What a check implementation returns; the runner stamps id/version/subject. */
+export interface CheckOutcome {
+  status: CheckStatus;
+  reason?: string;
+}
+
+/** Caller-provided inputs for one check invocation (e.g. the QC-20 probe path). */
+export type CheckInputs = Record<string, unknown>;
+
+/** Everything a check implementation may use. */
+export interface CheckContext {
+  subject: Subject;
+  params: Record<string, unknown>;
+  inputs: CheckInputs;
+  adapter: ProcessAdapter;
+}
+
+/** Outcome of running a shell probe through the thin process adapter. */
+export type AdapterOutcome =
+  | { ok: true; exitCode: number | null; stdout: string; stderr: string }
+  | { ok: false; kind: 'spawn-error' | 'timeout'; message: string };
+
+/** Request for the process adapter. */
+export interface AdapterRequest {
+  file: string;
+  args: string[];
+  cwd: string;
+  timeoutMs: number;
+}
+
+/**
+ * Thin process adapter: runs a command, owns NO verdict logic. Verdict parsing
+ * always lives in the check implementation (TS), never in the shell probe.
+ */
+export interface ProcessAdapter {
+  run(request: AdapterRequest): Promise<AdapterOutcome>;
+}
+
+/** Per-subject-kind check-set selection policy (versioned and digested). */
+export interface CheckSetPolicySpec {
+  version: string;
+  byKind: Record<SubjectKind, readonly string[]>;
+}
+
+export interface CheckSetPolicy extends CheckSetPolicySpec {
+  /** sha256 content digest of the policy spec. */
+  policyDigest: string;
+}
+
+/** Full typed evaluation report for one subject. */
+export interface EvaluationReport {
+  subject: Subject;
+  results: CheckResult[];
+  /** checkId → content digest of the definition that produced the verdicts. */
+  definitionDigests: Record<string, string>;
+  /** Version of the check-set policy used for subject selection. */
+  checkSetVersion: string;
+  state: AggregateState;
+}
+
+/** Options for `evaluateSubject`. */
+export interface EvaluateOptions {
+  subjectPath: string;
+  /** Restrict to these check ids; defaults to the subject kind's check set. */
+  checkIds?: string[];
+  /** Per-check inputs, keyed by check id (e.g. `{ 'qc-20-enforcement-verify': { probePath } }`). */
+  inputs?: Record<string, CheckInputs>;
+  /** Replace the built-in definitions (tests / future batches). */
+  definitions?: CheckDefinition[];
+  /** Inject a process adapter (tests / instrumentation). */
+  adapter?: ProcessAdapter;
+}

packages/quality-rails/src/index.ts

@@ -3,3 +3,35 @@ export * from './detect.js';
 export * from './scaffolder.js';
 export * from './templates.js';
 export * from './types.js';
+
+// RI-N4 evaluator (card RI-3-002): the public, programmatic entry points.
+export {
+  builtInDefinitions,
+  CHECK_SET_POLICY,
+  checkSetForKind,
+  defineCheck,
+  QC_19_RAILS_FILES_PRESENT,
+  QC_20_ENFORCEMENT_VERIFY,
+} from './evaluator/definitions.js';
+export { canonicalJson, digestContent, digestOfPolicy, digestOfSpec } from './evaluator/digest.js';
+export { createSpawnProcessAdapter } from './evaluator/adapter.js';
+export { aggregateState, detectSubjectKind, evaluateSubject } from './evaluator/runner.js';
+export type {
+  AdapterOutcome,
+  AdapterRequest,
+  AggregateState,
+  CheckContext,
+  CheckDefinition,
+  CheckDefinitionSpec,
+  CheckInputs,
+  CheckOutcome,
+  CheckResult,
+  CheckSetPolicy,
+  CheckSetPolicySpec,
+  CheckStatus,
+  EvaluateOptions,
+  EvaluationReport,
+  ProcessAdapter,
+  Subject,
+  SubjectKind,
+} from './evaluator/types.js';

scratchpads/ri-1-002-publish-gate-negative-controls.md

@@ -0,0 +1,45 @@
+# RI-1-002 — Publish-gate negative controls (SDLC-D-034 second half)
+
+- Task: RI-1-002 (docs/release-integrity workstream, PRD item RI-N1), issue ref #1275
+- Branch: `test/ri-050-publish-gate-negative` (base `origin/next` @ d8e0aec9 = PR #1277, RI-1-001)
+- Budget: worker estimate ~45K tokens; keep scoped to the two test files + scratchpad.
+
+## Objective
+
+Checked-in negative-control tests that PROVE the publish gate fails when it must:
+
+1. Broken mandatory check blocks every publish step (structural DAG proof from `.woodpecker/publish.yml`).
+2. Bypass shapes fail the checker: missing edge, hidden effect (non-`publish` name), detached verify, always-pass verify (`failure: ignore` / `success` override), conditional verify (`when`).
+3. Exact-commit identity: no HEAD-moving step between verify and publish effects; legitimate re-checkout requires verify to re-run after it.
+4. `verify-release.mjs` composition control: a SUBSET stage list fails the composition check.
+
+## Plan
+
+- NEW `scripts/publish-gate-structure.test.mjs` — self-contained structural checker (`assertPublishGateBlocksOnVerify`) + positive control on the real pipeline + one negative-control test per bypass shape (S1–S6, documented in file header) + positive control for the legitimate re-checkout shape.
+- EXTEND `scripts/verify-release.test.mjs` — refactor the stage-mirror test body into `assertStagesMirrorCi(stages, ci)`; add negative control dropping each stage one at a time (subset must throw).
+
+## Conventions confirmed
+
+- Root `test:checkout` = `node --test scripts/*.test.mjs` → new file auto-joins `pnpm test`.
+- Test-enumeration guard population is `*test*.sh` under `packages/mosaic/framework/tools/` only → unaffected.
+- Root eslint covers only `**/*.{ts,tsx}` → .mjs files need Prettier style only (printWidth 100, singleQuote, semi, trailingComma all).
+- Do NOT touch docs/TASKS.md, docs/release-integrity/TASKS.md, docs/scratchpads/.
+
+## Progress log
+
+- [x] Base verified: publish.yml `verify` step + verify-release.mjs present; HEAD contains origin/next.
+- [x] Wrote scripts/publish-gate-structure.test.mjs
+- [x] Extended scripts/verify-release.test.mjs (mirror fn + subset negative control)
+- [x] Gates: node --test scripts (31 tests pass), prettier clean on touched files, pnpm typecheck PASS, pnpm lint PASS, pnpm format:check PASS
+- [x] Committed ff585b88 + pushed, PR #1305 → next (no conflicts). Stopped before merge per task instruction.
+
+## Evidence
+
+- `node --test scripts/verify-release.test.mjs scripts/publish-gate-structure.test.mjs` → 31 tests, 0 fail.
+- Mutation sanity: temporarily removing the `verify` edge from build-gateway in publish.yml → structure test goes red (verified manually during dev, then reverted).
+- Gates run from repo root on this worktree; results in Progress log.
+
+## Risks / notes
+
+- Effect detection (`isPublishCommand`) is deliberately over-broad (any npm/pnpm/yarn command mentioning `publish`, any kaniko/docker-push/`--destination`) — fail-closed: a false positive forces justification, a false negative is the actual hazard.
+- `git fetch` flagged as HEAD-moving even though fetch alone doesn't move HEAD — fail-closed on the classic `fetch && reset` pair.

scratchpads/ri-4-001-prd-authority.md

@@ -0,0 +1,37 @@
+# Scratchpad — RI-4-001 One transitional PRD authority (RI-N3, #1275)
+
+- Objective: single PrdService authority in `@mosaicstack/prdy`; `mosaic prdy` and
+  `mission --plan` become thin adapters; mission↔PRD linkage persisted on disk;
+  Markdown export is a labeled generated view (never read back); import is
+  validated/conflict-aware with typed refusals.
+- Budget: ~35K tokens (card cap). Baselines: prdy build/lint rc=0, 0 tests;
+  mosaic build rc=0 (after root turbo build), lint rc=0, 1548 tests pass;
+  root build rc=0.
+- Plan: (1) extend store schema (version, missions linkage) (2) PrdService +
+  typed errors (3) wizard/cli route through service (4) mosaic adapters
+  (5) contract specs both packages (6) gates (7) sabotage control (8) report
+  to /var/tmp/ri-050/ri-4-001-report.md.
+- Decisions:
+  - Linkage lives ON the PRD document (`missions` array) — one authority file,
+    survives restart, no sidecar sync problems.
+  - `version` = content revision of sections/status (bumped by update/import
+    accept). Linkage writes bump `updatedAt` only, so ids/versions stay stable
+    for the card's "stable ids/versions" contract.
+  - Mission version marker = `mission.updatedAt` (gateway MissionInfo has no
+    numeric version field).
+  - Import reads YAML documents only — never the exported Markdown (keeps the
+    "no code path reads exported Markdown" invariant).
+  - Import of an existing id with identical core content → `identical` no-op;
+    divergent → typed `PrdImportConflictError` carrying proposed successor
+    (existing.version + 1, status draft, linkages preserved). Original bytes
+    untouched until explicit `acceptSuccessor`.
+  - `requirementIds` default `[]` at the mission command (no requirement
+    selection UI yet) — service accepts ids when a caller has them.
+- Progress log:
+  - [16:35] baselines captured (prdy 0 tests; mosaic 1548 after root build; root build rc=0)
+  - [16:38] store schema v2 + PrdService + wizard/cli rerouted; prdy build/lint green
+  - [16:40] mosaic adapters done; prdy spec 20/20 (found+fixed: import project-path leak, empty-store typed error, YAML timestamp coercion)
+  - [16:44] mosaic specs 9/9 (fixed commander from:'user' argv, vi.mock hoisting, restoreAllMocks wiping factory mocks)
+  - [16:45] all gates green; 4 commits (e291bfb, 2c5d208, a23826c, 540d6f1)
+  - [16:46] sabotage: linkage write removed → prdy 3 fail / mosaic 2 fail, 1548/1548 pre-existing pass; restored byte-identically; re-green 20/20 + 1557/1557
+  - [16:47] report written to /var/tmp/ri-050/ri-4-001-report.md — card complete

scripts/publish-gate-structure.test.mjs

@@ -0,0 +1,310 @@
+import assert from 'node:assert/strict';
+import { readFile } from 'node:fs/promises';
+import { createRequire } from 'node:module';
+import path from 'node:path';
+import test from 'node:test';
+
+// RI-1-002 / RI-N1 publish-gate NEGATIVE CONTROLS (SDLC-D-034).
+//
+// scripts/verify-release.test.mjs pins the POSITIVE structure of the publish
+// gate: every publish effect declares a direct `depends_on: verify` edge and
+// the verify step asserts commit identity + runs the canonical command. This
+// suite is the negative-control set: each test feeds a structural gate
+// checker a pipeline in which the gate is bypassed by ONE specific shape and
+// asserts the checker goes RED. The controls prove from the pipeline FILE —
+// never by executing Woodpecker — that a verify step that FAILS (nonzero
+// exit) blocks every publish effect.
+//
+// Woodpecker semantics these controls rely on:
+//   - A step that exits nonzero FAILS, and every step that transitively
+//     depends on a failed step is SKIPPED — never run. That skip is the only
+//     thing standing between a failed mandatory check and a publish effect.
+//   - `detach: true` removes the step from the wait graph: the pipeline does
+//     not wait for detached steps, so their failure can never block anything.
+//   - `failure: ignore` reports a failed step as success to the DAG.
+//   - `success: [codes...]` overrides which exit codes count as success;
+//     admitting any nonzero code launders a failed verification into green.
+//   - `when` on the verify step would skip verification entirely on some
+//     event/path classes while publish effects still run.
+//
+// Bypass shapes covered (one negative-control test each):
+//   S1  Missing edge — a publish effect whose dependency closure does not
+//       contain `verify` (a refactor drops the depends_on entry).
+//   S2  Hidden effect — a step whose NAME does not start with `publish` but
+//       whose COMMANDS publish npm packages or push images. Effects are
+//       classified by commands, so renaming a step cannot un-gate it.
+//   S3  Detached verify — `verify: { detach: true }`: publish steps no longer
+//       wait for verify, so the depends_on edge is decorative.
+//   S4  Always-pass verify — `failure: ignore`, or a `success` override
+//       admitting nonzero exit codes: verify fails, the DAG sees success.
+//   S5  Conditional verify — a `when`/path filter on verify itself.
+//   S6  Exact-commit drift — a HEAD-moving step (git checkout/switch/reset/
+//       clean/pull/clone/fetch) ordered between `verify` and a publish
+//       effect: the verified commit would not be the published commit. A
+//       LEGITIMATE re-checkout is allowed only when `verify` itself runs
+//       after it — positive control included.
+//   S7  Gate removal — the verify step deleted or renamed away entirely.
+
+// Reuse the monorepo's existing YAML parser (@mosaicstack/mosaic's direct
+// dependency) instead of adding a root dependency or vendoring a parser.
+const mosaicRequire = createRequire(
+  path.resolve(process.cwd(), 'packages', 'mosaic', 'package.json'),
+);
+const { parse: parseYaml } = mosaicRequire('yaml');
+
+const publishYmlPath = path.join(process.cwd(), '.woodpecker', 'publish.yml');
+
+async function readPublishPipeline() {
+  return parseYaml(await readFile(publishYmlPath, 'utf8'));
+}
+
+// A command has a publish EFFECT when it publishes npm packages (`publish`
+// anywhere after a package-manager token — `pnpm --filter "@x/*" publish`
+// puts flags and quoted filters between the binary and the subcommand) or
+// pushes an image (kaniko, docker push, or a registry --destination).
+// Deliberately over-broad: a false positive forces justification, a false
+// negative is the actual hazard.
+function isPublishCommand(command) {
+  return (
+    /(^|\s)\/kaniko\/executor\b/.test(command) ||
+    /(^|\s)docker\s+push\b/.test(command) ||
+    /(^|\s)--destination(\s|=)/.test(command) ||
+    (/\bpublish\b/.test(command) && /(^|\s)(npm|pnpm|yarn)(\s|$)/.test(command))
+  );
+}
+
+function hasPublishEffect(step) {
+  return (step.commands ?? []).some(isPublishCommand);
+}
+
+// A step is a publish effect when its name says so OR (S2) when any of its
+// commands does — classification must not depend on the name alone.
+function publishEffectSteps(pipeline) {
+  return Object.entries(pipeline.steps ?? {})
+    .filter(([name, step]) => name.startsWith('publish') || hasPublishEffect(step))
+    .map(([name]) => name);
+}
+
+// Transitive closure of a step's depends_on graph.
+function dependencyClosure(pipeline, stepName, seen = new Set()) {
+  const dependencies = pipeline.steps?.[stepName]?.depends_on ?? [];
+  for (const dependency of dependencies) {
+    if (seen.has(dependency)) continue;
+    seen.add(dependency);
+    dependencyClosure(pipeline, dependency, seen);
+  }
+  return seen;
+}
+
+// Deliberately over-broad: `git fetch` alone does not move HEAD, but the
+// classic re-checkout pair is `git fetch && git reset --hard <remote>`; a
+// fetch step sitting between verify and a publish effect deserves scrutiny,
+// so the gate fails closed on it.
+function movesHead(step) {
+  return (step.commands ?? []).some((command) =>
+    /(^|\s)git\s+(checkout|switch|reset|clean|pull|clone|fetch)\b/.test(command),
+  );
+}
+
+// The structural gate checker: green only when a failed (nonzero-exit)
+// verify provably blocks every publish effect on the same commit.
+function assertPublishGateBlocksOnVerify(pipeline) {
+  assert.ok(pipeline.steps, 'publish pipeline must define steps');
+  const verify = pipeline.steps.verify;
+  assert.ok(verify, 'publish pipeline must define a `verify` step (S7)');
+
+  // S5: a skipped verification authorizes publishes exactly as much as a
+  // failed one — verify must be unconditional.
+  assert.equal(verify.when, undefined, '`verify` must not carry a when/path filter (S5)');
+
+  // S3/S4: the depends_on edges are only meaningful if verify's own failure
+  // is both awaited and terminal for the DAG.
+  assert.equal(verify.detach, undefined, '`verify` must not be detached (S3)');
+  assert.equal(
+    verify.failure,
+    undefined,
+    '`verify` must not tolerate its own failure (S4: failure: ignore launders a failed gate into success)',
+  );
+  assert.equal(
+    verify.success,
+    undefined,
+    '`verify` must not override success exit codes (S4: nonzero codes would make failed verification pass)',
+  );
+
+  const effects = publishEffectSteps(pipeline);
+  assert.ok(effects.length > 0, 'publish pipeline must contain publish effect steps to guard');
+
+  const verifyClosure = dependencyClosure(pipeline, 'verify');
+  for (const stepName of effects) {
+    // S1: only the failure-skip semantics of the DAG stand between a failed
+    // verify and this effect — the verify edge in its closure is the proof.
+    const closure = dependencyClosure(pipeline, stepName);
+    assert.ok(
+      closure.has('verify'),
+      `publish effect '${stepName}' must transitively depend on verify (S1) — a failed verify must skip it`,
+    );
+
+    // S6: any step ordered after verify (outside its closure) but inside the
+    // effect's chain must not be able to move HEAD. If the pipeline
+    // legitimately re-checks-out, verify must run after the re-checkout.
+    for (const chainStep of closure) {
+      if (chainStep === 'verify' || verifyClosure.has(chainStep)) continue;
+      assert.ok(
+        !movesHead(pipeline.steps[chainStep]),
+        `step '${chainStep}' sits between verify and publish effect '${stepName}' and can move HEAD (S6)` +
+          ' — verify must re-run after any re-checkout',
+      );
+    }
+  }
+  return effects;
+}
+
+// A minimal but healthy gate used as the base for every negative-control
+// mutation: verify (identity + canonical command) → build → publish-npm,
+// with the publish effect blocked by verify both directly and through build.
+const HEALTHY_GATE_YAML = `
+steps:
+  verify:
+    image: node:24-alpine
+    commands:
+      - |
+        if [ -z "$CI_COMMIT_SHA" ] || [ "$CI_COMMIT_SHA" != "$(git rev-parse HEAD)" ]; then
+          echo "identity mismatch" >&2
+          exit 1
+        fi
+      - pnpm verify:release
+  build:
+    image: node:24-alpine
+    commands:
+      - pnpm build
+    depends_on:
+      - verify
+  publish-npm:
+    image: node:24-alpine
+    commands:
+      - npm publish
+    depends_on:
+      - build
+      - verify
+`;
+
+// Fresh parse per call so every negative control mutates its own object.
+function healthyPipeline() {
+  return parseYaml(HEALTHY_GATE_YAML);
+}
+
+test('the real publish pipeline: a failed verify provably blocks every publish effect', async () => {
+  const pipeline = await readPublishPipeline();
+  const effects = assertPublishGateBlocksOnVerify(pipeline);
+  assert.deepEqual(effects.sort(), [
+    'build-appservice',
+    'build-gateway',
+    'build-web',
+    'publish-next-npm',
+    'publish-npm',
+  ]);
+});
+
+test('fixture sanity: the healthy gate base passes the checker unmutated', () => {
+  assertPublishGateBlocksOnVerify(healthyPipeline());
+});
+
+test('S1 negative control: a publish effect with no verify edge fails the checker', () => {
+  const pipeline = healthyPipeline();
+  pipeline.steps['publish-npm'].depends_on = ['build'];
+  pipeline.steps.build.depends_on = [];
+  assert.throws(
+    () => assertPublishGateBlocksOnVerify(pipeline),
+    /publish-npm.*must transitively depend on verify/s,
+  );
+});
+
+test('S2 negative control: an npm publish hidden behind a non-publish step name fails the checker', () => {
+  const pipeline = healthyPipeline();
+  delete pipeline.steps['publish-npm'];
+  pipeline.steps.build.depends_on = [];
+  pipeline.steps.deploy = {
+    image: 'node:24-alpine',
+    commands: ['npm publish'],
+    depends_on: ['build'],
+  };
+  // Detection must be by COMMAND: the name says "deploy", the commands say
+  // publish — an un-gated effect under either reading.
+  assert.throws(
+    () => assertPublishGateBlocksOnVerify(pipeline),
+    /deploy.*must transitively depend on verify/s,
+  );
+});
+
+test('S2 negative control: a kaniko image push under a build-* name fails the checker when ungated', () => {
+  const pipeline = healthyPipeline();
+  delete pipeline.steps['publish-npm'];
+  pipeline.steps.build.depends_on = [];
+  pipeline.steps['push-platform-image'] = {
+    image: 'gcr.io/kaniko-project/executor:debug',
+    commands: ['/kaniko/executor --context . --destination reg.example/img:latest'],
+    depends_on: ['build'],
+  };
+  assert.throws(
+    () => assertPublishGateBlocksOnVerify(pipeline),
+    /push-platform-image.*must transitively depend on verify/s,
+  );
+});
+
+test('S3 negative control: a detached verify fails the checker', () => {
+  const pipeline = healthyPipeline();
+  pipeline.steps.verify.detach = true;
+  assert.throws(() => assertPublishGateBlocksOnVerify(pipeline), /detached \(S3\)/);
+});
+
+test('S4 negative control: failure: ignore on verify fails the checker', () => {
+  const pipeline = healthyPipeline();
+  pipeline.steps.verify.failure = 'ignore';
+  assert.throws(() => assertPublishGateBlocksOnVerify(pipeline), /tolerate its own failure/);
+});
+
+test('S4 negative control: a success override admitting nonzero exit codes fails the checker', () => {
+  const pipeline = healthyPipeline();
+  pipeline.steps.verify.success = [0, 1];
+  assert.throws(() => assertPublishGateBlocksOnVerify(pipeline), /success exit codes/);
+});
+
+test('S5 negative control: a when filter on verify fails the checker', () => {
+  const pipeline = healthyPipeline();
+  pipeline.steps.verify.when = [{ event: 'push' }];
+  assert.throws(() => assertPublishGateBlocksOnVerify(pipeline), /when\/path filter \(S5\)/);
+});
+
+test('S6 negative control: a HEAD-moving step between verify and publish fails the checker', () => {
+  const pipeline = healthyPipeline();
+  pipeline.steps.resync = {
+    image: 'node:24-alpine',
+    commands: ['git fetch origin', 'git reset --hard origin/main'],
+    depends_on: [],
+  };
+  pipeline.steps.build.depends_on = ['verify', 'resync'];
+  // resync sits AFTER verify in the publish chain (verify does not depend on
+  // it), so the verified commit could be replaced before publishing.
+  assert.throws(() => assertPublishGateBlocksOnVerify(pipeline), /resync.*can move HEAD/s);
+});
+
+test('S6 positive control: a legitimate re-checkout passes when verify re-runs after it', () => {
+  const pipeline = healthyPipeline();
+  pipeline.steps.resync = {
+    image: 'node:24-alpine',
+    commands: ['git fetch origin', 'git reset --hard origin/main'],
+    depends_on: [],
+  };
+  pipeline.steps.verify.depends_on = ['resync'];
+  pipeline.steps.build.depends_on = ['verify'];
+  // resync precedes verify in the chain, so verification covers the
+  // re-checked-out HEAD — the exact-commit contract holds.
+  assertPublishGateBlocksOnVerify(pipeline);
+});
+
+test('S7 negative control: deleting the verify step entirely fails the checker', () => {
+  const pipeline = healthyPipeline();
+  delete pipeline.steps.verify;
+  pipeline.steps['publish-npm'].depends_on = ['build'];
+  assert.throws(() => assertPublishGateBlocksOnVerify(pipeline), /`verify` step/);
+});

scripts/verify-release.mjs

@@ -22,6 +22,12 @@
 //   format        | format              | pnpm format:check
 //   test          | test                | pnpm test
 //   build         | publish.yml build   | pnpm build
+//   quality-rails | (canonical-only)    | the TS quality-rails evaluator
+//                |                     | (RI-N4, QC-19 monorepo subject). Like
+//                |                     | `build`, this stage has no ci.yml
+//                |                     | mirror; it is implemented by
+//                |                     | importing the evaluator CLI rather
+//                |                     | than duplicating its presence logic.
 //
 // Caller-provided prerequisites (kept at the pipeline level — see the comments
 // in .woodpecker/ci.yml): `bash` + `rsync` for the guard stages, `openssl` and
@@ -90,6 +96,15 @@ export const STAGES = [
     name: 'build',
     commands: ['pnpm build'],
   },
+  {
+    // RI-N4 (QC-19, card RI-3-002): the typed quality-rails evaluator, invoked
+    // as the implementation of the check it owns instead of a duplicated
+    // presence loop here. Canonical-only stage (no ci.yml mirror — same shape
+    // as `build`); runs AFTER build so the evaluator's dist/ exists. Subject
+    // is this repository (`.` → monorepo subject kind, per-subject check set).
+    name: 'quality-rails',
+    commands: ['node packages/quality-rails/dist/cli.js quality-rails evaluate --project .'],
+  },
 ];
 
 export function stageByName(name) {

scripts/verify-release.test.mjs

@@ -4,12 +4,15 @@ import { createRequire } from 'node:module';
 import path from 'node:path';
 import test from 'node:test';
 
-import { STAGES } from './verify-release.mjs';
+import { STAGES, stageByName } from './verify-release.mjs';
 
 // SDLC-D-034 checkout invariant: publication in .woodpecker/publish.yml is
 // bound to exact-commit terminal verification. This suite parses the real
 // pipeline files and fails red when the gate is bypassed, weakened, or drifts
-// out of sync with the canonical `pnpm verify:release` command.
+// out of sync with the canonical `pnpm verify:release` command. The negative
+// controls for pipeline DAG/bypass shapes live in
+// scripts/publish-gate-structure.test.mjs (RI-1-002); this file owns the
+// canonical-command composition controls.
 
 // Reuse the monorepo's existing YAML parser (@mosaicstack/mosaic's direct
 // dependency) instead of adding a root dependency or vendoring a parser.
@@ -219,14 +222,28 @@ steps:
   assert.throws(() => assertPublishGate(parseYaml(noIdentityPipeline)), /CI_COMMIT_SHA/);
 });
 
-test('the canonical verify:release stages mirror the PR CI pipeline one-for-one', async () => {
+// The composition check: the canonical stage table must mirror the PR CI
-  const ci = parseYaml(await readFile(ciYmlPath, 'utf8'));
+// pipeline's complete mandatory set. Parameterized by the stage list so the
-  const canonical = Object.fromEntries(STAGES.map((stage) => [stage.name, stage.commands]));
+// subset negative control below can prove a dropped stage goes red (RI-1-002:
+// the canonical command cannot silently lose a check).
+function assertStagesMirrorCi(stages, ci) {
+  const canonical = Object.fromEntries(stages.map((stage) => [stage.name, stage.commands]));
 
-  // The complete mandatory set, in gate order.
+  // The complete mandatory set, in gate order. `quality-rails` is a
+  // canonical-only stage (RI-N4, QC-19): like `build`, it has no ci.yml
+  // mirror to match — its contract is asserted separately below.
   assert.deepEqual(
-    STAGES.map((stage) => stage.name),
+    stages.map((stage) => stage.name),
-    ['sanitization', 'upgrade-guard', 'typecheck', 'lint', 'format', 'test', 'build'],
+    [
+      'sanitization',
+      'upgrade-guard',
+      'typecheck',
+      'lint',
+      'format',
+      'test',
+      'build',
+      'quality-rails',
+    ],
   );
 
   // Guard stages: ci.yml commands minus its `apk add` environment prep must be
@@ -269,9 +286,64 @@ test('the canonical verify:release stages mirror the PR CI pipeline one-for-one'
       `ci.yml test step must keep its pipeline-level prerequisite '${fragment}'`,
     );
   }
+}
+
+test('the canonical verify:release stages mirror the PR CI pipeline one-for-one', async () => {
+  const ci = parseYaml(await readFile(ciYmlPath, 'utf8'));
+  assertStagesMirrorCi(STAGES, ci);
+});
+
+test('a subset stage list fails the composition check — a dropped stage cannot pass silently', async () => {
+  const ci = parseYaml(await readFile(ciYmlPath, 'utf8'));
+  // Drop each stage one at a time: every stage is load-bearing, so every drop
+  // must go red. If any drop went green, a refactor could silently delete a
+  // mandatory check from the canonical command.
+  for (const stage of STAGES) {
+    const subset = STAGES.filter((entry) => entry.name !== stage.name);
+    assert.throws(
+      () => assertStagesMirrorCi(subset, ci),
+      Error,
+      `composition check must fail when the '${stage.name}' stage is dropped from the table`,
+    );
+  }
 });
 
 test('the root package.json exposes verify:release as the canonical command', async () => {
   const packageJson = JSON.parse(await readFile(path.join(process.cwd(), 'package.json'), 'utf8'));
   assert.match(packageJson.scripts['verify:release'], /scripts\/verify-release\.mjs/);
 });
+
+// RI-N4 (card RI-3-002): the `quality-rails` stage must route through the TS
+// evaluator instead of duplicating its presence logic inline. The evaluator
+// owns QC-19; this file keeps that delegation honest.
+function assertEvaluatorStage(stage) {
+  assert.ok(stage, 'canonical stages must include a quality-rails stage');
+  assert.ok(Array.isArray(stage.commands) && stage.commands.length > 0);
+  for (const command of stage.commands) {
+    assert.match(
+      command,
+      /packages\/quality-rails\/dist\/cli\.js.*quality-rails evaluate/,
+      `quality-rails stage command must invoke the evaluator CLI, got: '${command}'`,
+    );
+  }
+}
+
+test('the quality-rails stage invokes the evaluator rather than duplicating its logic', () => {
+  assertEvaluatorStage(stageByName('quality-rails'));
+});
+
+test('a quality-rails stage that re-implements presence logic inline fails the checker', () => {
+  // Negative control: replacing the evaluator invocation with an inline
+  // `test -f` presence loop is exactly the duplication RI-N4 forbids — the
+  // checker must go red on it.
+  const duplicated = {
+    name: 'quality-rails',
+    commands: ['test -f .husky/pre-commit && test -f .husky/pre-push'],
+  };
+  assert.throws(() => assertEvaluatorStage(duplicated), /must invoke the evaluator CLI/);
+});
+
+test('a quality-rails stage that silently drops the evaluator command fails the checker', () => {
+  const empty = { name: 'quality-rails', commands: [] };
+  assert.throws(() => assertEvaluatorStage(empty), /commands/);
+});

skills/mosaic-forge/SKILL.md

@@ -133,10 +133,13 @@ When the full `@mosaicstack/forge` package is available, Forge uses MACP task ex
 
 ```bash
 # Run from CLI
-mosaic forge run path/to/brief.md
+# Fails closed with a typed FORGE_NO_EXECUTOR capability error when no real
+# executor is wired — pass --simulate to opt into explicit typed simulation
+# (every result carries status `simulated`, which satisfies nothing).
+mosaic forge run path/to/brief.md [--simulate]
 
-# Resume interrupted run
+# Resume interrupted run (same fail-closed rule as forge run)
-mosaic forge resume .forge/runs/20260401-143022/
+mosaic forge resume .forge/runs/20260401-143022/ [--simulate]
 
 # Check status
 mosaic forge status .forge/runs/20260401-143022/