# Data Custody Contract — Pointer and Consent Schema (D14) Status: DRAFT — awaiting ratification (webui-audit S2, contract 7 of 9). Authority: PRD D14/D6 (Part I §7) — sensitive profile categories live in the **user's own brain ONLY**; PostgreSQL holds structural data, consent records, and pointers — never the content; "user data does not leak" is enforced by architecture, not policy. The Standalone split is a MAY with a recommended-default (kept for conversion forward-compatibility, D3). PRD D4 (Part I §6) — profile answers feed `USER.md` and/or the user's data store subject to the custody rule; connectors carry granular agentic-access consent. PRD D6 — estate brains hold operational records; only product-relevant material migrates into repository docs. Revision 2 (sol r1 findings F1–F6): the registry is now a full profile classifier that can represent non-sensitive rows, with an exact row shape, versioning, refusal semantics for unknown keys, and transition rules (F1); the consent model gains a concrete grantee reference model, an active-row uniqueness constraint, append-only re-grant semantics, and explicit mutation authority (F2); the pointer schema names its exact column set, binds the one-pointer rule to a database constraint, defines the `brain_ref` grammar and owner-bound resolution, replaces the unkeyed hash with a keyed construction, and bounds orphan repair (F3); the witnesses import the hierarchy contract's §6.2 column allowlist and §6.3-style closed route inventories, add negative controls and a column-type allowlist, and cover every binding rule (F4); the Standalone physical split, both valid layouts, the election record, and v1 phase timing are defined consistently with the wizard and mode-conversion contracts (F5); drafting additions are disclosed in §8 and the ruling request is one sentence with one decision (F6). Revision 3 (sol r2 findings F1–F9): reclassification gains a custody-config write fence (shared/exclusive row lock) so no concurrent writer can commit a stale-version route, a declared supported-version refusal, and witnesses for both transition directions (F1). The grantee reference becomes three per-type FK columns with a generated discriminant, real referent tables for `connector` and `feature` (contract-owned, empty at v1), the enrolled-agent binding named to the rank-4 family's table, and revoked-row immutability enforced by a database trigger (F2). The content hash is domain-separated per (user, category, locator) with canonical bytes, an embedded key id, and a rotation rule; the audit FK referent and the non-sensitive profile table (`profile_answers`) are named; tree-object resolution prohibits symlink escape (F3). The §7 route witnesses import hierarchy §6.3's full prong set, scan `plugins/`, extend the no-content probe to every platform table, and add the missing negatives (F4). `custody_config` is singleton by constraint (F5). §8 discloses the previously omitted policies (F6). Mutation authority is split into two disjoint actor predicates — subject mutations and the precondition-checked system auto-revocation (F7). Hash mismatch gets a representable pointer state with a defined terminal outcome (F8). Managed deletion is an ordered git-then-database protocol with fault-injection witnesses at both interruption points, replacing the impossible cross-store transaction (F9). Revision 4 (sol r3 findings F1/F2/F4/F6/F7/F9 residual, F10/F11 new): the write fence pins `READ COMMITTED` isolation and lock-first statement order for the reclassification migration, and the supported-version declaration becomes a named source symbol checked under the same lock as each decision (F1). The user foreign keys adopt the identity store's actual key type — `users.id` is `text` (BetterAuth identifiers), so every `user_id` column is `text` (F10). The agent referent binds to the live platform `agents` table by symbol, since `identity-lifecycle.md` defines no enrollment surface; an amendment re-binds if one arrives (F2). The system auto-revocation predicate is withdrawn entirely: grantee cessation is handled structurally by `ON DELETE CASCADE` from the grantee referent (as subject cessation already is from `users`), default-deny ends access, and the audit event table keeps the durable history — which removes the trigger/actor contradiction rather than patching it (F2, F7). Sensitive-write routes gain a static write-set closure witness so a derived representation cannot land in any generic platform table (F4). Managed deletion step 2 becomes a compare-and-delete on the hash removed in step 1, witnessed against a concurrent rewrite (F9). Ordinary sensitive writes get an ordered content-first protocol with deterministic locators, ack-after-pointer-commit, idempotent retry, and fault-injection witnesses (F11). §8 discloses every rule above plus the previously undisclosed §4.3 committed-tree/symlink rule (F6). Revision 5 (sol r4 findings F9/F11 residual, F12/F13 new): the pointer row gains a monotonic `generation` counter bumped by every upsert, and managed-deletion step 2 conditions on the generation recorded in step 1 instead of the content hash, closing the identical-content ABA in which a byte-identical rewrite reproduced the old hash and a stale deletion unpointed fresh content (F9). The ordered-write protocol gains a contract-defined locator scheme — injective over (category, slot), with the slot set fixed to the single value `answer` at version 1 — and a per-locator advisory lock spanning both steps, so same-locator writers are serialized and a completed write always leaves the pointer hash describing the current blob; §7.16 adds the concurrent-writer, crash-before-acknowledgment, stale-retry, and locator-injectivity witnesses (F11). Cascaded consent deletion now emits a semantic cessation event in the same transaction as the referent deletion, written by a trigger on the consent table, so the audit trail records when and why each active grant ended, not only that it began (F12). The stale §3.6/§5.2 cross-references and the "first predicate" remnant are corrected (F13). Revision 6 (sol r5 residuals F9/F12 + F14): the deletion and repair token becomes the PAIR (row `id`, `generation`) — the uuid primary key is minted fresh on every insert and never reused, so it is the row-incarnation identity the bare counter lacked; a stale compare-and-delete can no longer match a row reincarnated at generation 1 after an orphan-repair delete, and §7.10 adds the delete/reinsert ABA control (F9). The §3.7 locator lock now covers EVERY blob/pointer mutator for a locator, not only ordinary writers: managed deletion (§4.7) and reconciliation repair (§4.5) acquire the same advisory lock across their observe-and-mutate spans, so the reverse interleaving — a deleter destroying a blob a locked writer has committed but not yet pointed — is unschedulable while the tokens keep the crash-window (lock released by session death) harmless; §7.10 adds the reverse interleaving and its lock-removed control (F14). The cessation-event snapshot gains the deleted consent row's `id` and its `user_id`, so events from two subjects' grants to one ceasing grantee are attributable to their exact rows and subjects, witnessed with a two-subject case (F12). Revision 7 (sol r6 F15): the reconciliation try-acquire no longer discharges a locator for the cycle — a failed try-acquire QUEUES the locator within the same cycle, and before the cycle completes the sweep revisits every queued locator with a blocking acquire, running the same locked observation and token-conditioned repair. A triggered reconciliation cycle is complete only when every in-scope pointer has been processed under its locator lock, so the §4.5 one-cycle orphan bound holds unconditionally on lock contention; the retry terminates because every §3.7 hold is transaction- or session-scoped and session end releases the lock automatically. §7.10 adds the lock-held contention control (sweep pass skips without blocking; cycle completion waits for the queued repair; the orphan is deleted before the cycle reports completion), its repeated-contention variant, and a third static control failing an implementation that defers a skipped locator to a later cycle (F15). Revision 8 (sol r7 residual F15): the one-cycle bound gains its missing termination proof. §3.7 adds two liveness rules for the locator lock: a declared finite HOLD DEADLINE, enforced with session-level timeout mechanisms on every lock-holding path — a live holder that stalls past the deadline is terminated, session end releases the lock, and the stalled-holder schedule collapses into the already-harmless crash window — and an arrival-order GRANT rule, so a blocked waiter's wait is bounded by the current holder and the waiters already ahead, never by later arrivals. §4.5's blocking retry therefore terminates within a wall-clock bound fixed when it starts waiting, and the implementing PR derives and declares a maximum cycle duration whose exceedance is surfaced as an error, never a silent hang. §7.10 adds the hold-deadline control (a non-cooperating stalled holder is terminated at the deadline and the orphan is still deleted before cycle completion), the grant-order control (a stream of later contenders cannot starve the queued repair, and the declared cycle-duration limit is exercised as a checked bound), and a fourth static control failing any lock-holding path without the declared deadline (F15). Revision 9 (sol r8 residual F15 + new F16): the hold deadline becomes independently enforceable and the external store gains a fence. §3.7's deadline is now enforced by a deadline supervisor whose liveness does not depend on the stalled holder — the database's timeout mechanisms are defense in depth, not the enforcement path, since step 1's brain commit is not a database statement — and every brain mutation performed under a locator lock is FENCED: applied conditionally on the locator's brain state still equaling the value recorded at lock acquisition, so a terminated holder's in-flight external operation, resuming after the lock has passed on, is refused atomically and can never remove or replace a later holder's committed content (F16). §4.5's deferred repair now RESERVES its queue place at the moment its try-acquire fails, fixing the predecessor set at skip time and closing the window between skip and queue entry; the cycle maximum becomes the computed formula `M = S + Σᵢ(D·(Qᵢ+1)+R)` over snapshots taken at or before each reservation, fixed when the last reservation is placed and never enlarged; and exceedance is an explicit transition — the cycle reports failure, never completion, withdraws its pending reservations, and persists a durable carry-over set the next cycle processes first — so the orphan bound binds every completing cycle and every non-completing path is a surfaced error, and the grant-order rule must be bound to a documented mechanism guarantee or an application FIFO queue, not assumed (F15). §7.10 adds two external-step deadline controls (a stalled managed deletion and a stalled ordinary-writer brain commit, each terminated by the supervisor mid-external-operation, resumed after a fresh holder's acknowledged write, and asserted refused by the fence), the reservation control, the exceedance control, and static controls binding the grant-order guarantee and the fenced deletion path; §7.16 adds the static fenced-commit control (F15, F16). Revision 10 (sol r9 residual F16 + new F17): the brain fence becomes ABA-resistant. §3.7's fence value is now a per-locator fence EPOCH with two invariants: every successful brain mutation mints a fresh, never-reused value — byte-identical recommits and transitions to absence included, absence being a unique tombstone state rather than a recurrable bare-missing-path value — and every holder performs a FENCE-ADVANCE immediately after lock acquisition, minting a fresh epoch before its first brain observation or mutation, so every earlier holder's recorded fence is invalidated at handoff, before the new holder relies on exclusivity. A stale external tail is therefore refused unconditionally: neither a byte-identical recommit, nor a return to absence, nor resumption before the new holder's first mutation can re-present a consumed epoch (F16). The live holder's fence-pass now follows from handoff invalidation rather than from lock possession, and a defect-only refusal has a specified abort-and-surface transition. §4.5 repair performs the advance before the absence observation it relies on. The unqualified distinct-locator no-contention claim is narrowed to locator locks: a shared atomic update point may cost swap-loop retries across locators but never invalidates a distinct locator's condition — a liveness cost, never a correctness coupling (F17). §7.10 adds two ABA fence controls (byte-identical stale deletion; absence→write→delete→stale-writer resurrection), a handoff-invalidation control with a repair-observation variant, and a static fence-epoch-invariant control; §7.16's static fenced-commit control is restated over the epoch (F16, F17). Revision 11 (sol r10 residual F16 + new F18): the fence-advance itself is fenced against an older acquisition. §3.7 adds a per-locator ACQUISITION TOKEN minted by the lock authority at every lock grant — fresh, strictly increasing, existing before the holder's first brain operation and carried unchanged through every brain operation and swap-loop retry of that acquisition — and the brain store records the greatest accepted token atomically with the locator's epoch and content state, refusing any brain mutation, the fence-advance included, that carries an older token. The advance's acceptance condition is exactly this comparison (an epoch condition would be circular, an unconditional mint would let a terminated holder's delayed advance supersede a later holder's); a refused advance mints nothing and returns no usable fence, with a mandatory abort-release-surface transition that is an EXPECTED outcome after deadline supersession, not a defect. Revision 10's "refused unconditionally" claim is precisified to an exact boundary: after the newer holder's advance linearizes, every older-acquisition operation — its delayed advance included — is refused; a stale content mutation linearizing BEFORE that advance serializes entirely before anything the new holder relies on (the advance preserves the content state it finds and mints after it), indistinguishable from completion before the previous holder's lock release — the defined legal outcome. The pre-step-1 interruption predicates in §3.7, §4.7, and the §7.16 fault injection are restated exactly — content presence, content bytes, and database rows unchanged, the advance's epoch mint and token recording expressly permitted — resolving their contradiction with the mandatory advance (F18). §7.10 adds a stale-advance refusal control (fresh-writer and repair-observation variants) and a pre-advance linearization control, and extends the fence-invariant static control with the token invariants; §7.16's static fenced-commit control carries the token (F16, F18). Revision 12 (sol r11 minors F18 residual, F19, F20): three exactness corrections to the revision-11 mechanism, no behavioral change. §3.7's token comparison is made exact at both ends: strictly less-than refuses (an already-recorded token re-presented by a swap-loop retry or a later mutation of the same acquisition is accepted), and an absent recorded token compares lower than every minted token, so a virgin locator's first advance is accepted and atomically creates the token record — with a §7.10 virgin-locator control and a delayed-first-grant race variant (F19). §7.10's remaining managed-deletion fault predicate ("both stores unchanged") is restated as the exact §4.7 predicate — content presence, content bytes, and database rows unchanged, the advance's epoch mint and token recording expressly permitted (F18 closed). The pre-advance-linearization outcome no longer assigns the (`id`, `generation`) pair to every stale operation: the §3.7 boundary text and the §7.10 control now distinguish the database halves — deletion's and repair's step 2 stays pair-conditioned, while a terminated ordinary writer performs no pointer upsert and can continue only through a fresh acquisition under a new token (F20). This contract binds the profile-category registry (§2), the custody placement rule (§3), the pointer schema (§4), the consent schema and its evaluation (§5), mode application (§6), witnesses (§7), and disclosed drafting additions (§8). It defines schemas and placement; wizard step flow stays with contract 3 (`onboarding-wizard.md`), mode and conversion with contract 6 (`mode-conversion.md`), identity with `identity-lifecycle.md`, tool mapping with `tool-gateway-mapping.md`. ## 1. Definitions 1. **User brain**: the git-tracked per-user data store. Its two valid Standalone layouts are defined in §6.2; in Enterprise it is the user's own brain repository. 2. **Sensitive content**: any profile answer or derived text in a §2 category classified `sensitive`. 3. **Pointer**: a database record referencing sensitive content that lives in a user brain, carrying no content (§4). 4. **Grantee**: a non-subject principal that may be granted access to a user's sensitive content: a platform agent (a row of the `agents` table, §5.2), a connector, or a platform feature (§5.2). 5. **Consent record**: a database record granting one grantee access to one category of one user's data (§5). 6. **Registry version**: the monotonically increasing integer identifying the active state of the §2 registry. 7. **Physical split** (Standalone): sensitive user content living in a per-user brain repository separate from the estate mosaic-brain, as opposed to the unsplit layout where it lives in a dedicated user-files subtree of the single mosaic-brain (§6.2). 8. **Audit event table**: the single platform table into which the hierarchy contract (contract 1, `hierarchy-schema.md` §5.2) commits its semantic audit events. No sibling contract names the physical table, so this contract binds it by identity, not by name: every `audit_event_id` column in §3.5, §4.1, §5.1, and §6.3 is a foreign key to that table, the implementing PR that creates the audit store binds these FKs to it, and the §7.1 witness asserts that each such FK references the same physical table the contract-1 audit write path inserts into. ## 2. Profile-category registry 1. **Model.** The registry is the single classifier for every profile category, closed and versioned, in the platform database. Table `profile_category_registry`, columns exactly: | Column | Type | Constraints | | ---------------- | ----------- | ---------------------------------------------------- | | `category_key` | text | primary key | | `classification` | text | NOT NULL, CHECK in (`sensitive`, `non-sensitive`) | | `since_version` | integer | NOT NULL (registry version that introduced this row) | | `created_at` | timestamptz | NOT NULL | The current registry version is a single integer held in `custody_config` (§6.3). Rows are added or reclassified ONLY by amendment to this contract shipped as a migration that bumps the registry version; no runtime write path may insert, update, or delete registry rows. 2. **Initial registry (version 1).** Drawn from the D4 profile step and D14's "disabilities, family, communication style, and similar": | Category key | Classification | Covers | | ------------------------- | -------------------------- | ------------------------------------------------------------------- | | `disabilities` | sensitive | disabilities including ADHD/autism/PDA/vision | | `family-social` | sensitive | family, pets, friends | | `communication-style` | sensitive | desired agent communication style, voice-matching interview product | | `personal-interests` | sensitive | hobbies, likes/dislikes | | `connector-content` | sensitive | email and drive content reached through user connectors | | `professional-background` | non-sensitive (per ruling) | professional background summary used for agent configuration | | `education` | non-sensitive (per ruling) | education summary used for agent configuration | The `Covers` column is contract documentation, not a database column. Account identity fields (email, name, credentials) are identity-contract data, not profile custody data, and have no registry row. 3. **Unknown category — refusal.** A profile write naming a `category_key` with no registry row is REFUSED with an explicit error. Nothing is stored anywhere, no registry row is auto-added (§2.1), and no pointer is created (so the §4.1 foreign key is never asked to reference a missing row). Fail-closed means refusal, not silent routing. 4. **Reclassification transitions.** A reclassification ships as a contract amendment plus migration that bumps the registry version. Non-sensitive → sensitive: the same migration moves every existing relational value for that category into its owning user's brain, creates the pointers, and deletes the relational values, all before the new version activates. Sensitive → non-sensitive: existing brain content and pointers remain valid and are never automatically materialized into the database; only writes evaluated after the new version activates route relationally. 5. **Unreadable registry — refusal.** If the registry or its version cannot be read at decision time, every routing and consent decision that depends on it is refused. There is no cached-default or assume-sensitive fallback that performs a write. 6. **Write fence.** Every profile-write transaction reads the current registry version by taking a **shared lock** on the `custody_config` row (`SELECT ... FOR SHARE`) inside the same database transaction that performs the write, and routes by the version so read. A reclassification migration takes an **exclusive lock** on that row (`SELECT ... FOR UPDATE`) before its scan and holds it through the value moves and the version bump. Isolation and statement order are part of the fence, not implementation freedom: both the profile-write transaction and the migration transaction run at `READ COMMITTED`, and the migration's `SELECT ... FOR UPDATE` is the FIRST statement of its transaction — no read precedes it, so the migration establishes no snapshot before it holds the lock. Under `READ COMMITTED` each subsequent statement takes a fresh snapshot, so the migration's scan — which runs only after the lock is granted — sees every row committed by every `FOR SHARE` holder that finished before the lock was granted. Consequently a concurrent writer either commits before the lock is granted — its value is then visible to and swept by the scan — or blocks until the migration commits and then routes by the new version. No interleaving exists in which a value routed under the old version escapes the scan, including the writer-commits-while-migration- waits interleaving (witness §7.6g). 7. **Supported version.** Each release declares the exact registry version its code implements as an exported compile-time constant in the custody module of the shared schema package; the implementing PR names the symbol and §7.6(h) pins it by symbol identity, so the declaration is source, not configuration. Every profile routing decision and every consent decision reads `custody_config.registry_version` under a `SELECT ... FOR SHARE` on the `custody_config` row inside the same database transaction as the decision (write decisions share the §2.6 fence transaction; read-only consent evaluations take the shared lock in their own transaction), so the supported-version comparison and the decision it gates use one atomically read version. If the declared version does not equal the version so read, the decision is refused — a readable but unsupported registry is a refusal, never best-effort routing under either version (witness §7.6h). ## 3. Custody placement rule 1. Sensitive content is written to the owning user's brain ONLY. PostgreSQL tables MUST NOT store sensitive content in any column — not as text, not as excerpts or previews, not as encodings, and not as embeddings or other derived representations that reconstruct content. 2. The database MAY hold, about sensitive content: the pointer records of §4, the consent records of §5, the registry of §2, and the custody configuration of §6.3. Nothing else. 3. Every write path for profile answers routes by the registry: `sensitive` → brain write + pointer upsert; `non-sensitive` → the `profile_answers` table (§3.5); unknown → refusal (§2.3). The routing decision is made server-side from the registry at its current version, inside the §2.6 write fence; a client-supplied classification or routing override is ignored. 4. D6 boundary: operational records stay in estate brains and are linked, not migrated. This contract governs user-profile custody only and creates no new obligation on estate brains. 5. **Non-sensitive profile table.** Non-sensitive profile answers land in exactly one declared table, `profile_answers`, columns exactly: `id` (uuid, primary key), `user_id` (text, NOT NULL, FK → users(id) ON DELETE CASCADE), `category_key` (text, NOT NULL, FK → profile_category_registry(category_key)), `value` (text, NOT NULL), `created_at` and `updated_at` (timestamptz, NOT NULL), and `audit_event_id` (uuid, NOT NULL, FK → the §1.8 audit event table); plus UNIQUE (`user_id`, `category_key`). A value may sit in this table ONLY while its category's registry classification is `non-sensitive` (§2.4 migrates the rows on reclassification). No other platform table stores profile answer content. 6. **User-key type binding.** The identity store declares `users.id` as `text` (BetterAuth identifiers, `packages/db` schema); every `user_id` column in this contract (§3.5, §4.1, §5.1) is therefore `text`, matching the referenced key's declared type exactly. The identifiers are opaque strings, not guaranteed UUIDs; no custody column re-types, parses, or reformats them. If a sibling contract ever migrates the user key type, an amendment to this contract follows it (§7.1 asserts the FK types match the live referenced columns). 7. **Ordered sensitive-write protocol.** An ordinary sensitive write spans two stores and is ordered, content first. **Locator scheme.** The locator is a pure function `L(category_key, slot) → brain_ref` declared by the implementing PR, with three contract-bound properties: it is deterministic (a retry of the same write resolves to the same path), it is INJECTIVE — distinct (category_key, slot) pairs map to distinct `brain_ref` values, witnessed with a collision control (§7.16) — and it is stable across releases (a scheme change is a contract amendment with a migration). An **answer slot** is the stable identifier of one answer's position within a category; at version 1 every registry category is single-valued (mirroring §3.5's UNIQUE (`user_id`, `category_key`)), so the slot set is the single constant `answer`. An amendment introducing a multi-valued category must extend the slot set while preserving injectivity. **Serialization.** The entire write — both steps — runs under a per-locator mutual exclusion: before step 1 the writer acquires an exclusive advisory lock keyed on (`user_id`, `category_key`, slot) (e.g. a session-scoped `pg_advisory_lock` on a stable hash of the triple), held until step 2 commits or the writer's session ends (session end releases it automatically, so a crashed holder never wedges the locator). Two liveness rules bound this exclusion. **Hold deadline:** every session that acquires a locator lock — ordinary writer, managed deleter (§4.7), reconciliation repair (§4.5) — runs under a declared finite hold deadline: the implementing PR declares the value and enforces it through a **deadline supervisor** whose liveness is INDEPENDENT of the holder — a separate live component, not a timer inside the possibly-stalled holder process, that observes lock acquisition times and terminates the database session of any holder past its deadline. The database's session-level timeout mechanisms (statement timeout, idle-in-transaction timeout) are configured as defense in depth but are not the enforcement path: a statement timeout aborts a statement without ending the session, an idle-in-transaction timeout fires only on an idle open transaction, and step 1's brain commit is not a database statement at all — so a session stalled inside an external operation is terminated by the supervisor, which no database-side mechanism replaces. A holder that stalls past the deadline thus has its session TERMINATED — session end then releases the lock automatically per the rule above. The terminated holder's partial DATABASE protocol is exactly the crash window the tokens already make harmless (§4.5, §4.7); its possibly still in-flight EXTERNAL operation is made harmless by the brain fence below. No holder — live, stalled, or crashed — holds a locator lock longer than the declared deadline. **Grant order:** conflicting waiters on one locator lock are granted in arrival order — a requester that starts waiting after another never acquires before it. The implementing PR MUST bind this property to a mechanism, not assume it: either it cites the pinned advisory-lock mechanism's documented queue-order grant guarantee, or, absent such a documented guarantee, it routes every conflicting locator-lock acquisition through an application-level FIFO queue whose order invariant is itself part of the implementation; the binding is checked by a §7.10 static control and the property is exercised dynamically (§7.10) — so a blocked waiter's wait is bounded by the current holder's remaining deadline plus the deadline-bounded holds of the waiters already ahead of it, never by later arrivals. Writers to one locator are therefore totally ordered; writes to distinct locators do not contend FOR LOCATOR LOCKS. The brain store's atomic update point MAY be shared across locators (a single head compare-and-swap in a git-backed store serializes physically unrelated writes): losing that swap costs a retry of the swap loop, and the retry re-validates only the writer's own per-locator condition (Brain fence below), which an unrelated locator's mutation never changes — so cross-locator contention at a shared update point is a liveness cost, never a correctness coupling. Because no two same-locator writes interleave, a completed write always leaves the pointer's `content_hash` describing the blob currently committed at the locator — the divergent interleaving (writer A's pointer over writer B's content) is unschedulable, not merely unlikely. **Brain fence.** Deadline termination releases the lock while an external brain operation started by the terminated holder may still be in flight; lock release is therefore not, by itself, a fence for the external store. Two per-locator values fence it: an **acquisition token** ordering holders and a **fence epoch** ordering mutations. The acquisition token is minted by the lock authority at EVERY grant of the locator lock: a fresh, strictly increasing per-locator value handed to the holder with the grant — it therefore exists before the holder's first brain operation, independently of any brain state — and carried UNCHANGED through every brain operation of that acquisition, including every retry of a swap loop; a holder never re-mints or refreshes its token. The fence epoch is a per-locator value stored in the brain store, atomically readable and comparable together with the locator's content state, whose defining property is that every change to it MINTS a fresh value never before used at that locator. A per-locator monotonic counter satisfies this, as does a unique commit identity in a git-backed store; the locator's content BYTES do not — two byte-identical states at different times are distinct epochs — and ABSENCE is itself an epoch-bearing state, recorded as a unique tombstone, never a bare missing path whose "value" can recur. Three rules bind them. **Minting:** every successful brain mutation under a locator lock — a content commit, a byte-identical recommit, a content deletion (transition to absence) — advances the locator's epoch to a fresh never-reused value, atomically with the mutation. **Token recording:** the brain store records, per locator and atomically with the locator's epoch and content state, the greatest acquisition token it has accepted; EVERY brain mutation under a locator lock — the fence-advance below included — presents its holder's acquisition token, is REFUSED atomically when that token is less than the recorded value, and records it on acceptance. The comparison is exact at both ends: it is strictly less-than — re-presenting the already-recorded token, as a swap-loop retry or a later mutation of the same acquisition does, is accepted — and a locator with NO recorded token compares as lower than every minted token, so a virgin locator's first presented token is never refused by this comparison and its acceptance atomically CREATES the token record; an implementation whose absent-token comparison can refuse or fail to record the first holder is nonconforming (witness §7.10). **Fence-advance at acquisition:** immediately after acquiring the locator lock, before its first brain observation or mutation at the locator, the holder performs a fence-advance — an atomic epoch mint that leaves the locator's content presence and bytes unchanged — and records the resulting epoch as its **fence value**; the advance invalidates every fence recorded by any earlier holder before the new holder relies on anything it reads or on its exclusivity. The advance's OWN acceptance condition is exactly the token comparison — not the epoch (no fence exists for the acquisition until the advance returns, so an epoch condition would be circular) and not nothing (an unconditional mint would let a terminated holder's delayed advance supersede a later holder's): the token is minted at lock grant, so it exists before the advance, and grants for one locator are totally ordered, so the comparison encodes acquisition order. A refused advance returns NO usable fence: the holder MUST NOT perform any brain observation or mutation, MUST abort its protocol run before any later step, release the lock if it still holds it, and surface the error — and this refusal is an EXPECTED outcome for an acquisition whose lock was deadline-released and re-granted before its delayed advance applied, not a defect. Every other brain mutation performed under a locator lock — an ordinary write's step-1 content commit (below), a managed deletion's step-1 content deletion (§4.7) — is applied conditionally: the brain store applies it only if the locator's epoch at application time still equals the holder's recorded fence value AND the token-recording comparison accepts, with the checks and the application atomic with respect to other brain mutations (compare-and-swap semantics; an atomic ref update in a git-backed store satisfies this, with the per-locator conditions re-validated inside the swap loop when the swap is taken on a coarser head). A stale operation — one whose session was terminated and whose lock has passed to a newer holder — is refused at an exact boundary. Once the newer holder's fence-advance has linearized, EVERY operation of the older acquisition is refused: its epoch condition fails (the advance consumed its epoch, and no sequence of mutations can re-present a consumed one — whether the intervening history changed the bytes, recommitted byte-identical content, returned the locator to absence, or consists of the advance alone), its token comparison fails, and its own delayed fence-advance is refused by the token comparison, minting nothing — a superseded acquisition cannot manufacture a fresh fence. Before the newer advance linearizes, a stale content mutation whose conditions still hold MAY linearize first: the advance then finds the post-mutation state, expressly preserves the content presence and bytes it finds, mints its epoch after that mutation, and the new holder's first relied-on observation follows its own advance — so the stale effect serializes entirely before anything the new holder relies on, indistinguishable from the same operation completing before the previous holder's lock release, an outcome the protocol already defines as legal. What the stale operation's DATABASE half may then do depends on its protocol: a managed deletion's or repair's step 2 remains conditioned on the recorded (`id`, `generation`) pair (§4.5, §4.7) and cannot unpoint any later row incarnation; an ordinary write's terminated run performs NO pointer upsert — its database session is dead, the committed-but-unpointed content falls into the existing step-1-to-step-2 interruption window repaired by §4.5, and any retry is a NEW lock acquisition under a fresh token, never a continuation of the dead run. These are the only two outcomes: a stale operation either serializes wholly before the new holder's advance or is refused; it never applies after the advance, and it cannot remove or replace content committed by any later holder, whether it resumes microseconds or hours after the termination. A live holder's own fenced mutation cannot fail its conditions — not because it holds the lock, but because handoff invalidation plus minting mean the only epochs minted since its fence-advance are its own, and because lock grants are serialized by the lock itself, no greater acquisition token can be recorded while it holds the lock; a holder whose fenced mutation is nonetheless refused (reachable only under a defect breaching these invariants) MUST abort its protocol run, release the lock, and surface the error — it never retries the mutation under its stale fence. Every brain-mutation path carries the fence and the acquisition token, and every brain-RELYING observation path (reconciliation repair's absence observation, §4.5) performs its fence-advance before the observation it relies on; a path applying an unfenced or token-free brain mutation fails the §7.10 and §7.16 static controls. **Steps.** Step 1 commits the content to the owning user's brain at `L(category_key, slot)` as a fenced mutation (Brain fence above); step 2, only after step 1 has committed, upserts the pointer row (insert, or update of the existing row for that (user, category, brain_ref), incrementing `generation` per §4.1) in its own database transaction inside the §2.6 fence. The write is acknowledged to the caller only after step 2 commits. Interruption before step 1 leaves content presence, content bytes, and every database row unchanged — the post-acquisition fence-advance may already have minted an epoch and recorded an acquisition token, the only brain change the protocol permits before step 1. Interruption between the steps leaves committed brain content with no pointer: unpointed content is inert — it lives in the correct custody store, no read path serves it (reads resolve through pointers only), and nothing dangles in the database. Interruption after step 2 commits but before the acknowledgment reaches the caller leaves the converged state; the retry below observes it and returns success without corrupting it. The client's retry — the write was never acknowledged — re-acquires the locator lock, recommits the same locator, and completes the pointer upsert, converging to the pointed state; the retry is idempotent whichever interruption point preceded it. A stale retry (an earlier write retried after a later same-locator write completed) also executes under the lock and leaves a consistent pointed state — same-locator outcomes are last-completed-write wins, and no completion order can leave the hash describing a non-current blob. The reverse order (pointer before content) is forbidden: an ordinary write never creates a pointer whose content has not committed. Witness §7.16 fault-injects the interruption points, drives retry convergence, the same-locator concurrent interleaving, the stale retry, and the injectivity control. ## 4. Pointer schema 1. **Exact columns.** Table `profile_pointers`, columns exactly: | Column | Type | Constraints | | ---------------- | ----------- | ------------------------------------------------------ | | `id` | uuid | primary key | | `user_id` | text | NOT NULL, FK → users(id) ON DELETE CASCADE (§3.6) | | `category_key` | text | NOT NULL, FK → profile_category_registry(category_key) | | `brain_ref` | text | NOT NULL, CHECK against the §4.3 grammar | | `content_hash` | text | NOT NULL (§4.4 construction) | | `generation` | bigint | NOT NULL (1 on insert; every §3.7 upsert increments) | | `created_at` | timestamptz | NOT NULL | | `updated_at` | timestamptz | NOT NULL | | `mismatch_at` | timestamptz | NULL unless the pointer is in the §4.6 mismatch state | | `audit_event_id` | uuid | NOT NULL, FK → the §1.8 audit event table | plus the database constraint UNIQUE (`user_id`, `category_key`, `brain_ref`) — the one-pointer rule is a constraint, not a convention. 2. **Opacity.** `brain_ref` and every other pointer column MUST NOT embed content or content-derived text (no titles, snippets, or free-text descriptions). A locator is structural, not descriptive. 3. **`brain_ref` grammar and owner binding.** `brain_ref` is a normalized repository-relative POSIX path: one or more segments matching `[A-Za-z0-9][A-Za-z0-9._-]*`, joined by `/`, with no leading `/`, no empty segment, and no `.` or `..` segment; the stored form matches `^[A-Za-z0-9][A-Za-z0-9._-]*(/[A-Za-z0-9][A-Za-z0-9._-]*)*$`. Resolution ALWAYS roots at the brain owned by the row's `user_id` (the resolver takes the owner from the row, never from the locator); the locator carries no repository, host, or user component, so a cross-user or traversal reference is unrepresentable, not merely forbidden. Resolution operates on the brain repository's committed git tree (tree and blob objects), never through filesystem path lookup: if any segment of the path resolves to a symbolic link — or to anything other than a tree (intermediate segments) or a blob (final segment) — resolution is refused. A symlink therefore cannot redirect a locator outside the owner's brain (witness §7.8). 4. **`content_hash` construction.** `content_hash` is `hmac-sha256::` — HMAC-SHA-256, keyed with a platform integrity key held in the secrets backend and never stored in the database or any repository, over the length-prefixed concatenation of: a fixed domain-separation string naming this contract and the construction version, the row's `user_id`, the row's `category_key`, the `brain_ref`, and the canonical content bytes. The canonical content bytes are the exact bytes of the committed blob the locator resolves to (§4.3) — no normalization. Because the input is domain-separated per (user, category, locator), equal answers from different users or categories store different hashes: the column is neither an offline dictionary oracle (external key) nor an equality/correlation oracle across rows (domain separation). `` names the key used; rotation introduces a new key id for new writes, re-verification accepts any still-registered key id, and a key is retired only when no stored hash references it. On read, a hash mismatch refuses the read and puts the pointer into the §4.6 mismatch state. 5. **Bounded orphan repair.** Pointers are deleted when their content is deleted; dangling pointers are repaired toward deletion, never toward re-creating content in the database. Reconciliation for a user's pointers runs on two triggers: every profile write for that user, and a periodic sweep whose interval the implementing PR declares (at most daily). A pointer whose content is absent is deleted by the next triggered reconciliation cycle that reports COMPLETION — no orphan survives a completed cycle; a cycle that cannot complete within its computed maximum fails as an explicit surfaced error with durable carry-over (below), never silently — and repair performs no database content write. Per-pointer repair runs under the §3.7 locator lock: before observing content absence for a pointer, reconciliation acquires the pointer's locator lock, performs the §3.7 fence-advance under repair's own acquisition token — invalidating any stale in-flight operation's fence BEFORE the observation repair relies on, so no terminated holder's resuming mutation — its delayed fence-advance included, refused by the §3.7 token comparison — can change the locator between repair's observation and its DELETE — holds the lock across the observation and the DELETE, and releases it after. The sweep pass MAY use a try-acquire so it never blocks mid-pass behind a writer, but a failed try-acquire defers the locator, never discharges it — and the deferral RESERVES its place: at the moment the try-acquire fails, the sweep submits the locator's blocking lock request (the **reservation point**), so the request enters the lock queue immediately and its predecessor set — the current holder plus the waiters already queued at that instant — is fixed at skip time; contenders arriving afterward are later arrivals that §3.7 grant order places behind it, and there is no window between the try-acquire failure and queue entry in which a new contender can join the predecessor set. The sweep proceeds with the pass while the reservation waits, and before the cycle completes it returns to every reserved locator, completes the blocking acquire, and runs the same locked observation and token-conditioned repair. A triggered reconciliation cycle is COMPLETE only when every in-scope pointer has been processed under its locator lock; deferral moves work to the end of the cycle, never into a later one except through the exceedance transition below — which fails the cycle loudly, never completes it — so the completed-cycle bound above holds unconditionally on lock contention. The blocking retry terminates within a WALL-CLOCK BOUND, not merely eventually: every holder is subject to the §3.7 hold deadline (a stalled live session is terminated by the independent supervisor at the deadline and session end releases the lock — voluntary release is not assumed), and grants follow §3.7 arrival order (no later requester overtakes the queued reservation), so the reservation is granted within the current holder's remaining deadline plus the deadline-bounded holds of the waiters already queued ahead — a finite quantity fixed at its reservation point, independent of how many contenders arrive later. Cycle completion is therefore itself bounded by a COMPUTED maximum fixed within the cycle: the implementing PR declares the hold deadline `D`, a sweep-pass bound `S` over the cycle's scope (the scope is snapshotted when the cycle starts), and a per-locator repair bound `R` (at most `D`), and computes the cycle maximum `M = S + Σᵢ (D·(Qᵢ+1) + R)` over the reserved locators `i`, where `Qᵢ` is the number of waiters queued ahead of reservation `i` at its reservation point. Every quantity in the formula is a snapshot taken at or before the corresponding reservation, so `M` is fully determined when the last reservation is placed and is NEVER enlarged afterward — queue growth after a reservation cannot extend it, because later arrivals are not predecessors. **Exceedance transition.** A cycle that reaches its computed `M` with reserved locators unprocessed stops: it withdraws its pending lock reservations, reports FAILURE — never completion — surfaces the error, and PERSISTS its unprocessed reserved locators as a durable CARRY-OVER set. The next triggered reconciliation cycle takes the carry-over set into its scope FIRST — fresh reservations, its own computed maximum — before its own sweep. An orphan is deleted before its cycle reports completion, within that cycle's computed `M`; the only path past a cycle boundary is the exceedance transition, an explicit surfaced failure with durable carry-over — never a silent hang and never a silent drop. The repair DELETE is additionally token-conditioned like §4.7 step 2: it records the row's `id` and `generation` when it observes the content absent and conditions the DELETE on BOTH — the uuid `id` is minted per insert and never reused, so it identifies the row incarnation; `generation` orders upserts within the incarnation. A concurrent §3.7 write that recommits the locator between the observation and the delete (bumping `generation`), and a delete-then-reinsert that reincarnates the locator's row (fresh `id`, `generation` back at 1), each make the repair match zero rows — reconciliation never unpoints content it did not observe absent, under either token dimension. 6. **Mismatch state.** A failed §4.4 verification stamps the pointer's `mismatch_at` and the read is refused; every subsequent read of a pointer with `mismatch_at` set is refused without re-serving content. Reconciliation (§4.5 triggers) re-verifies each flagged pointer: verification success clears `mismatch_at`; content absent → the pointer is deleted (§4.5); persistent mismatch → the pointer is retained flagged with reads refused — the terminal outcome. The subject's next successful profile write for that category replaces the content, recomputes the hash, and clears the flag. Mismatch handling never deletes brain content and never copies content into the database. 7. **Managed deletion protocol.** A managed deletion of sensitive content spans two stores and is an ordered protocol, not a single transaction. **Lock coverage**: the deleter FIRST acquires the §3.7 advisory lock on the target locator (`user_id`, `category_key`, slot) — the same lock ordinary writers hold — before reading the pointer row or touching either store, and holds it until step 2 commits or the deleter's session ends. Every blob/pointer mutator for one locator — ordinary writes (§3.7), managed deletion, and reconciliation repair (§4.5) — therefore serializes on one lock: the reverse interleaving, in which a deleter reads the old pointer and destroys a blob that a lock-holding writer has committed in its step 1 but not yet pointed, is unschedulable, because the deleter cannot enter its protocol while the writer holds the lock. **Steps**: step 1 commits the content deletion to the user's brain repository as a FENCED mutation (§3.7 Brain fence — conditioned on the fence epoch recorded at the deleter's post-acquisition fence-advance, carrying the deleter's acquisition token, and itself minting the locator's unique absence tombstone epoch), recording the target row's `id` AND `generation` as read after lock acquisition and before step 1; step 2, only after step 1 has committed, deletes the pointer row in its own database transaction as a **compare-and-delete** — the DELETE is conditioned on the recorded `id` AND `generation`. The token is this PAIR, not the content hash and not the generation alone, deliberately: every §3.7 upsert increments `generation` even when it recommits byte-identical content (whose §4.4 hash is unchanged), so a same-row rewrite moves the row past the recorded generation; and the uuid `id` is minted fresh on every insert and never reused, so a row REINCARNATION — the orphan row deleted by §4.5 repair, then an ordinary write recreating the locator with a fresh row whose `generation` is back at 1 — cannot reproduce the recorded pair even though the counter value repeats. In every such case step 2 matches zero rows, deletes nothing, and the fresh content stays pointed: deletion never removes a pointer row it did not observe. The lock excludes the live interleavings; the pair token makes the crash windows harmless — a deleter whose session dies after step 1 releases the lock (§3.7 rule), and whatever runs afterward, its never-run step 2 has no effect and a later deletion request is a NEW protocol run reading current state. A deleter TERMINATED at the §3.7 hold deadline while its step-1 brain deletion is still in flight is likewise harmless on the brain side: the in-flight deletion is fenced, so once a later holder performs its §3.7 fence-advance — before that holder's first mutation, and regardless of whether its eventual commit is byte-identical to the state the deleter recorded — the stale deletion's conditional application fails and removes nothing: content committed and acknowledged after the termination cannot be destroyed by the terminated deleter's resuming operation (witness §7.10). Interruption before step 1 commits leaves content presence, content bytes, and every database row unchanged — the deleter's post-acquisition fence-advance may already have minted an epoch and recorded an acquisition token, the only brain change the protocol permits before step 1. Interruption between the steps leaves a dangling pointer, which §4.5 repairs toward deletion within one reconciliation cycle. At no point does any compensation write content into the database (witness §7.10, including both concurrent-rewrite interleavings — changed content and identical content — the delete/reinsert reincarnation, and the reverse writer/deleter interleaving with its lock-removed control). ## 5. Consent schema and evaluation 1. **Exact columns.** Table `profile_consents`, columns exactly: | Column | Type | Constraints | | ---------------- | ----------- | ----------------------------------------------------------------------------- | | `id` | uuid | primary key | | `user_id` | text | NOT NULL, FK → users(id) ON DELETE CASCADE (the data subject, §3.6) | | `grantee_type` | text | NOT NULL, CHECK in (`agent`, `connector`, `feature`) | | `agent_id` | uuid | NULL, FK → agents(id) ON DELETE CASCADE (§5.2) | | `connector_id` | text | NULL, FK → custody_connector_registry(connector_key) ON DELETE CASCADE (§5.2) | | `feature_key` | text | NULL, FK → custody_feature_registry(feature_key) ON DELETE CASCADE (§5.2) | | `grantee_ref` | text | generated stored: COALESCE(agent_id::text, connector_id, feature_key) | | `category_key` | text | NOT NULL, FK → profile_category_registry(category_key) | | `state` | text | NOT NULL, CHECK in (`granted`, `revoked`) | | `granted_at` | timestamptz | NOT NULL | | `revoked_at` | timestamptz | CHECK ((state = 'granted') = (revoked_at IS NULL)) | | `actor` | text | NOT NULL (the authenticated principal that recorded the change) | | `audit_event_id` | uuid | NOT NULL, FK → the §1.8 audit event table | plus: the CHECK that exactly one referent column is non-NULL and matches the discriminant — `(grantee_type = 'agent') = (agent_id IS NOT NULL)` AND `(grantee_type = 'connector') = (connector_id IS NOT NULL)` AND `(grantee_type = 'feature') = (feature_key IS NOT NULL)` — and the partial unique index UNIQUE (`user_id`, `grantee_type`, `grantee_ref`, `category_key`) WHERE `state = 'granted'` — at most one active grant per (user, concrete grantee, category), as a database constraint. 2. **Grantee reference model.** A grantee is identified by (`grantee_type`, the matching referent column); its canonical display form is `:`. The discriminant set is closed at the three CHECK values, and every referent is a real foreign key: - `agent` → `agent_id` references the platform **`agents` table** (`agents.id`, uuid) as declared in the shared schema package (`packages/db`). This is a binding to the live schema, named honestly: the ratified `identity-lifecycle.md` defines no agent enrollment surface or enrolled-agent table, so there is no sibling-defined referent to bind to — the `agents` table is today's only platform representation of agents, and the §7.11 witness pins the FK's referenced table to that schema symbol. If a sibling amendment later introduces an agent-enrollment surface with its own table, an amendment to this contract re-binds the FK; until then, rows in `agents` are the closed set of possible agent grantees. - `connector` → `connector_id` references this contract's own table `custody_connector_registry`, columns exactly: `connector_key` (text, primary key), `since_version` (integer, NOT NULL), `created_at` (timestamptz, NOT NULL). Rows are added ONLY by amendment to this contract (§2.1 pattern); the table is EMPTY at version 1. - `feature` → `feature_key` references this contract's own table `custody_feature_registry`, with the same three-column shape (`feature_key` primary key) and the same amendment-only rule; EMPTY at version 1 (no feature grantee exists until an amendment names one). A consent row naming a nonexistent referent violates its FK and is refused by the database. A grant to one agent confers nothing on another agent of the same type; the constraint key includes `grantee_ref`, so two same-type grantees are distinct rows. 3. **Default deny.** Absence of an active `granted` row for (user, grantee, category) means no access. There are no implicit grants, no platform-admin bypass, and no mode in which default-deny is suspended. 4. **Mutation authority — subject only.** Every consent mutation (create, grant, revoke) must satisfy the single predicate: the authenticated actor IS the row's data subject — the server asserts `actor` equals the principal of the row's `user_id` before the mutation commits. A platform admin has no consent-mutation capability over another user's rows — an admin grant or revocation on another user's data is refused at write time, closing the write-side route around §5.3. There is no system-actor mutation predicate: revision 3's system auto-revocation is withdrawn (disclosed, §8), because grantee cessation is handled structurally instead — **Cessation by cascade**: when a grantee ceases to exist — the `agents` row is deleted, or a `custody_connector_registry` / `custody_feature_registry` row is removed by an amendment migration — the referent FK's `ON DELETE CASCADE` removes that grantee's consent rows in the same transaction as the referent deletion. Access ends with the rows (§5.3 default deny: no row, no access); no UPDATE is performed, so no actor question arises. **Cessation event.** The cascade is not silent: an AFTER DELETE trigger on `profile_consents` inserts, for every deleted row whose deletion the §5.5 trigger permitted (i.e. a cessation cascade), one semantic **consent-cessation event** into the §1.8 audit event table, in the same database transaction as the referent deletion — if the deleting transaction rolls back, no cessation event survives, and if it commits, the events commit with it. Each event carries: the cause (`subject-cessation` or `grantee-cessation`, plus the parent table and deleted key), the initiating actor — a platform surface that deletes a referent sets a transaction-local actor variable the trigger reads; absent one, the event records actor `system:cascade` (a migration or direct referential action) — an immutable snapshot of the ended grant identifying its exact row and subject: the deleted consent row's `id`, its `user_id` (the data subject), grantee type and ref, `category_key`, `state`, `granted_at`, and `revoked_at` — and the event time. The row `id` and `user_id` are REQUIRED precisely because the cause alone identifies only the ceasing referent: when two users hold grants to one ceasing grantee, the cascade removes both rows in one transaction, and only the per-row snapshot distinguishes which subject's grant each event records. The snapshot is consent metadata, never profile content, so §3.1 is not implicated. The audit table therefore records both ends of every grant's life: the mutation events written at grant/revoke time survive the cascade (audit events are never cascade-deleted), and the cessation event records when, why, and by whom the active grant ended. The same rule already governs subject cessation via the `user_id` FK. §7.12 witnesses the subject predicate, its refusal complement, the cascade, and the cessation event's atomicity, fields, and survival; every §7.3-enumerated consent-mutation route asserts the subject predicate. 5. **Revocation and re-grant.** Revocation flips exactly one active row to `revoked` and stamps `revoked_at`; it is effective for every access evaluated after the revoking write commits. Revoked rows are retained as history and never mutated again — enforced by a database trigger on `profile_consents` that permits UPDATE only when it is the granted → revoked transition (the state flip plus the `revoked_at` stamp, every other column unchanged), and permits DELETE only when the row's referenced subject or grantee no longer exists — the trigger checks that the `users` row named by `OLD.user_id`, or the referent row named by the grantee column, is absent, which is true exactly during a §5.4 cessation cascade (referential cascades delete the parent row before the dependent rows, so the check distinguishes a cascade from a standalone DELETE). Every other UPDATE and every DELETE with both referents alive is rejected, so a revoked row cannot be flipped back to `granted` and history cannot be pruned by anyone, including through direct SQL (witness §7.13). A re-grant after revocation inserts a NEW row (append-only history) — repeated grant/revoke cycles are represented as successive rows, and the §5.1 partial unique index guarantees the old revoked rows cannot keep access live. 6. **Evaluation placement.** Access to sensitive content is mediated by the platform (Gateway/tooling) evaluating consent before any brain read on behalf of a grantee; the evaluation fails closed (`rbac-grant-model.md` §3.5 pattern), including when the consent state cannot be read (§2.5 pattern). The user reading their own data is not a grantee and needs no consent row. 7. Consent records govern agentic/feature access to user data. They are distinct from hierarchy grants (contract 2) and confer no platform authorization. ## 6. Mode application 1. The §2–§5 schemas are mode-independent: Standalone and Enterprise use the same tables and the same routing rule. 2. **Standalone layouts.** The D14 physical split remains a MAY. Its two valid layouts are: **split** — sensitive user content in a per-user brain repository separate from the estate mosaic-brain (the recommended default); **unsplit** — sensitive user content in the dedicated user-files subtree `users//` of the single mosaic-brain. §3 binds the logical user-brain region identically in both layouts; the layout election changes where the region lives, never whether routing applies. 3. **Election record.** The election lives in this contract's own one-row table `custody_config`, columns exactly: `id` (uuid, primary key), `singleton` (boolean, NOT NULL, DEFAULT true, CHECK (`singleton`), UNIQUE — the one-row rule as a database constraint: a second row cannot satisfy the CHECK and the UNIQUE index simultaneously), `standalone_layout` (text, NOT NULL, CHECK in (`split`, `unsplit`), default `split`), `registry_version` (integer, NOT NULL, §2.1), `elected_at` (timestamptz, NOT NULL), `actor` (text, NOT NULL), `audit_event_id` (uuid, NOT NULL, FK → the §1.8 audit event table). It is written at bootstrap and amended only by an explicit operator action; contract 6's exact, immutable mode record is not touched or extended by this contract. The row is also the §2.6 fence anchor. 4. **Phase timing.** v1 ships the §2–§6 schemas and the D14 database boundary, and the wizard collects no sensitive category in v1 (contract 3 §3), so v1 contains no sensitive write surface. §3 binds every sensitive write path from the moment one exists — the first profile surface that accepts a sensitive category (P2/P3) activates routing, consistent with contract 6 §5's v1 slice (D14 database boundary only, custody mechanics outside v1). 5. **Conversion precondition.** An operator electing `unsplit` accepts conversion-time partitioning: `mode-conversion.md` §4.2 requires the per-user partition to exist before the Enterprise flip, so conversion from an unsplit install performs the partitioning first. 6. In Enterprise, the split is mandatory (D3 table); no-leakage between users is enforced by §3 placement plus §5 default-deny — there is no cross-user read path to sensitive content through the database, because the database has no content to serve. ## 7. Verification requirements Binding on the implementing PRs. Every witness below MUST name, in its implementation, the exact tables, columns, commands, and source roots it scans; "the custody tables" means `profile_category_registry`, `profile_pointers`, `profile_consents`, `custody_config`, `custody_connector_registry`, and `custody_feature_registry`; "the declared profile table" means `profile_answers` (§3.5). 1. **Column-allowlist witness** (hierarchy contract §6.2 style): the custody tables' live column sets are exactly §2.1/§4.1/§5.1/§5.2/§6.3 and `profile_answers` is exactly §3.5; no platform table outside `profile_answers` carries profile answer content; every `audit_event_id` FK (§3.5, §4.1, §5.1, §6.3) references the same physical table that the contract-1 (hierarchy §5.2) audit write path inserts into (§1.8 binding); and every FK column's declared type equals the referenced column's live declared type — in particular each `user_id` is `text` matching `users.id` (§3.6 binding) and `agent_id` is `uuid` matching `agents.id` (§5.2 binding). 2. **Column-type allowlist witness:** the custody tables and `profile_answers` use only the column types named in §2.1/§3.5/§4.1/§5.1/§5.2/§6.3 (uuid, text, integer, bigint, timestamptz, boolean) — no bytea, json/jsonb, array, vector, or tsvector column exists in them, closing the encoded/derived-representation routes by type rather than by probe alone. 3. **Closed write-route witness** (hierarchy contract §6.3, full prong set): a static, re-export-aware inventory over `apps/`, `packages/`, and `plugins/` (production code, tests excluded) enumerates every module that writes the custody tables, `profile_answers`, or profile answers generally, detecting access through each of hierarchy §6.3's prongs — schema-symbol imports, SQL string literals naming the tables, raw-execution primitives, and runtime code construction — with the database client reachable only through a closed importer allowlist. Every enumerated write route implements §3.3 registry routing inside the §2.6 fence, and every enumerated consent-mutation route asserts the §5.4 subject predicate; a route outside the enumeration, or a client import outside the allowlist, fails the assertion. **Write-set closure for sensitive paths:** for every enumerated route that performs a sensitive write, the same static inventory (same prong set) enumerates every database table that route's code path can write; the set MUST equal exactly {`profile_pointers`, the §1.8 audit event table} — the brain write goes through the declared brain-write surface, never the database. Any additional platform-table write statically reachable from a sensitive-write path — an embedding insert, a cache row, any derived representation in any generic table — fails the witness. Control: a mutated route that adds one insert into an arbitrary existing platform table is detected. This closes the §3.1 escape in which a conforming-looking route also writes derived sensitive data elsewhere: §7.5 probes for planted content; this closure bounds what a sensitive-write route can write at all. 4. **Closed brain-read witness** (same style): the inventory enumerates every production route that reads user-brain content on behalf of a grantee, and every enumerated route calls the §5.6 consent evaluation; a brain-read route outside the enumeration fails. 5. **Routing witness:** a sensitive-category answer submitted through the profile surface results in a brain write plus a pointer row and zero content bytes in the database; a non-sensitive answer lands in its declared table. The no-content probe is a negative control: the witness first plants the fixture text in a scratch column of a throwaway table to prove the probe detects it, then asserts its absence — as plaintext, base64, hex, and JSON-string encodings — across every column of every table in the platform database: §3.1 bans sensitive content in every column of every table, so the probe scope is the entire schema, not only the custody tables. 6. **Registry witnesses:** (a) the version-1 registry state is exactly the seven §2.2 rows with their classifications; (b) a `professional-background` answer routes relationally (or per the ruling's alternative); (c) an unknown `category_key` is refused with nothing stored (§2.3); (d) with the registry unreadable, the write is refused (§2.5); (e) a runtime insert/update/delete against `profile_category_registry` outside a migration is refused (§2.1); (f) a reclassification migration (non-sensitive → sensitive) on seeded data moves the values to brains, creates pointers, and leaves zero relational values (§2.4); (g) fence race: a profile write transaction opened before a reclassification migration takes its exclusive lock either commits before the lock is granted (its value is swept) or blocks and, on commit, routes by the new version — the witness drives both interleavings, INCLUDING the interleaving in which the writer holds `FOR SHARE` when the migration requests `FOR UPDATE`, the migration waits, the writer commits its relational row, and the migration's scan (running after lock grant, at `READ COMMITTED`, with the lock as its first statement per §2.6) is asserted to sweep that row — and asserts no relational value for the reclassified category exists after any interleaving completes (§2.6); the witness also asserts the migration implementation takes its exclusive lock as the first statement of its transaction (a migration variant that reads before locking fails review of this witness's static check); (h) with `custody_config.registry_version` set to a value the release's declared compile-time constant (§2.7, pinned by symbol) does not equal, profile writes and consent decisions are refused, and the witness asserts the version comparison reads the row under the §2.7 shared lock in the same transaction as the refused decision (§2.7); (i) sensitive → non-sensitive: after the version bump, existing brain content and pointers remain intact and readable, nothing is materialized into the database, and the next write for that category lands in `profile_answers` (§2.4). 7. **Server-side classification witness:** a client-supplied classification or routing override on a profile write is ignored; the registry decision is applied (§3.3). 8. **Pointer-constraint witnesses:** inserting a second pointer for the same (user, category, brain_ref) violates the §4.1 unique constraint; a `brain_ref` failing the §4.3 grammar (leading `/`, `..` segment, empty segment) is rejected by the CHECK; resolution of a valid `brain_ref` under user A's row never reads user B's brain (owner binding, §4.3); with a symlink committed at a locator's path or as an intermediate segment, resolution is refused and no content outside the owner's brain is read (§4.3 tree-object rule). 9. **Hash witnesses:** `content_hash` verifies via the keyed §4.4 construction; the database value alone, without the external key, does not equal any unkeyed digest of the fixture content (oracle control); the same fixture content stored for two users, and for two categories of one user, yields different stored hashes (domain-separation control — no cross-row equality oracle); after a key rotation, a pointer written under the old key id still verifies and a new write stores the new key id; a mismatch refuses the read and stamps `mismatch_at`, subsequent reads stay refused, reconciliation of a persistently mismatched pointer retains it flagged without deleting brain content or writing content to the database, and the subject's re-write clears the flag (§4.4, §4.6). 10. **Orphan and deletion-protocol witnesses:** starting from a PRE-EXISTING orphan (content already absent, pointer present), the next triggered reconciliation deletes the pointer and writes no content anywhere in the database (§4.5). The §4.7 protocol is fault-injected at both interruption points: killed before the brain commit, content presence, content bytes, and every database row are unchanged — the deleter's post-acquisition fence-advance may already have minted an epoch and recorded an acquisition token, the only brain change the protocol permits before step 1; killed between the brain commit and the pointer delete, the dangling pointer is deleted by the next reconciliation — and at no point is content written to the database or restored to the brain. **Concurrent-rewrite interleavings (two):** deletion step 1 commits and the deleting session is fault-injected dead before step 2 — the §4.7 lock releases with the session, opening the crash window; a concurrent §3.7 write then recreates the same `brain_ref` and completes its pointer upsert; the stale step-2 compare-and-delete is then replayed with the recorded token. The witness drives this once with CHANGED content (new `content_hash`, incremented `generation`) and once with BYTE-IDENTICAL content (unchanged `content_hash`, incremented `generation` — the ABA case a hash token would miss), and in both asserts the DELETE matches zero rows, the fresh pointer survives, its content resolves and verifies, and no unpointed content and no dangling pointer exist afterward. **Delete/reinsert reincarnation (third interleaving):** in the same crash window, §4.5 reconciliation first removes the now-orphan pointer row; an ordinary §3.7 write then recreates the locator as a FRESH row — new `id`, `generation` back at 1 and therefore EQUAL to the recorded generation value; the replayed stale step 2 is asserted to match ZERO rows (the recorded `id` names the dead incarnation), and the fresh pointer survives with its content resolving and verifying — the case a bare generation token fails. **Reverse writer/deleter interleaving:** a §3.7 writer holds the locator lock and has committed its step-1 brain content but not its pointer upsert when a managed deletion for the same locator starts. The witness asserts the deletion BLOCKS at §4.7 lock acquisition until the writer's pointer upsert and release, and that afterward no acknowledged write's content is absent and no pointer names absent content; a lock-removed control schedules the deleter's observation and content removal inside the writer's step-1-to-step-2 gap and asserts the witness DETECTS the resulting acknowledged-write loss, establishing that the lock is what excludes it. **Lock-held contention control (one-cycle bound):** starting from a PRE-EXISTING orphan, a §3.7 writer session holds the orphan's locator lock while the triggered reconciliation's sweep pass runs. The witness asserts the pass completes without blocking and without repairing the held locator (the try-acquire skip), that the cycle does NOT report completion while the lock is held, and that after the holder releases, the queued blocking retry deletes the orphan under the recorded (`id`, `generation`) token BEFORE the cycle reports completion — the orphan is gone by cycle completion, proving the §4.5 bound under contention. A repeated-contention variant hands the lock to a second holder the moment the first releases and asserts the queued locator is retried until acquired — the orphan is still deleted by cycle completion once the last holder releases, never dropped after one failed attempt. A **hold-deadline control** starts from a pre-existing orphan, has a live holder acquire the orphan's locator lock and stall past the declared §3.7 hold deadline WITHOUT ending its session (idle transaction kept open), and asserts the holder's session is terminated at the deadline, the lock releases without any voluntary action by the holder, and the queued blocking retry then deletes the orphan under the recorded token before the cycle reports completion — proving termination does not depend on a cooperative holder. A **grant-order control** blocks the reserved repair behind a holder, keeps a stream of later contenders requesting the same locator lock while the reservation waits, and asserts the reservation is granted before every later contender and the cycle completes within its computed maximum `M` — proving later arrivals cannot starve the queued repair, and exercising the computed cycle maximum as a checked bound rather than prose. A **reservation control** has the sweep's try-acquire fail against a holder and asserts the blocking lock request is enqueued AT the try-acquire failure (the §4.5 reservation point), not at end of pass — then starts contenders immediately after the failed try-acquire and asserts every one is granted after the reservation, and that the cycle's `M` was fixed when the last reservation was placed and is not enlarged while the late contenders queue. Two **external-step deadline controls** exercise termination at the real external boundary, not an idle transaction. First, the stalled managed deletion: a §4.7 deleter acquires the locator lock and starts its step-1 brain deletion, which is suspended mid-operation — the session neither idle in a transaction nor executing a database statement; the control asserts the independent §3.7 supervisor terminates the session at the deadline with no cooperation from the holder and the lock releases; a fresh §3.7 writer then acquires the lock, commits new content at the locator with its pointer upsert, and is acknowledged; the suspended stale deletion is then RESUMED and asserted REFUSED by the brain fence — the fresh content is still present, its pointer resolves and verifies, and no mismatch is stamped. Second, the stalled ordinary writer: a §3.7 writer stalls inside its step-1 brain commit, is terminated at the deadline, a fresh writer completes a full acknowledged write at the same locator, and the stale commit is resumed and asserted refused — the locator serves the fresh writer's bytes and the fresh pointer's hash verifies. Two **ABA fence controls** exercise value recurrence, the cases a state-equality fence misses. Byte-identical stale deletion: a §4.7 deleter records its fence, is suspended inside its step-1 brain deletion and terminated by the supervisor; a fresh §3.7 writer then recommits BYTE-IDENTICAL content (unchanged §4.4 hash) with its pointer upsert and is acknowledged; the stale deletion is resumed and asserted REFUSED — the identical bytes carry a fresh epoch — with the content still present, resolving, and verifying, and no mismatch stamped. Absence resurrection: an ordinary writer records the locator's ABSENCE as its fence, is suspended inside step 1 and terminated; a fresh writer completes an acknowledged write at the locator; a managed deletion then removes that content and its pointer, returning the locator to absence; the stale writer is resumed and asserted REFUSED — the post-deletion absence is a fresh tombstone epoch, not a recurrence of the recorded one — with the locator still absent and NO unpointed content existing anywhere afterward. A **handoff-invalidation control**: a holder is terminated mid-external-operation; the next holder acquires the lock and performs its §3.7 fence-advance but has NOT yet applied any brain mutation when the stale operation is resumed — the stale operation is asserted refused (the advance alone invalidated it, before any fresh mutation), and the new holder's own subsequent fenced mutation is asserted to succeed. A repair-observation variant runs the same schedule with reconciliation as the new holder: repair advances, observes content absence, and the stale writer resuming before repair's DELETE is asserted refused — the DELETE then completes under its recorded (`id`, `generation`) token and NO unpointed content is resurrected at the locator. A **stale-advance refusal control** exercises the §3.7 token guard on the advance itself: writer W1 acquires the locator lock and its mandatory fence-advance is suspended BEFORE its atomic application; the independent supervisor terminates W1's session at the deadline and the lock releases; the next holder acquires the lock (receiving a greater acquisition token) and completes its own fence-advance; W1's suspended advance is then RESUMED and asserted REFUSED by the token comparison — it mints no epoch and returns no usable fence — and W1's protocol run is asserted to abort with the error surfaced, performing no further brain observation or mutation, while the next holder's fence remains valid and its subsequent fenced mutation is asserted to succeed. A fresh-writer variant runs the next holder as an ordinary §3.7 writer through a full acknowledged write; a repair-observation variant runs it as §4.5 reconciliation — repair advances, observes content absence, W1's late advance resumed before the DELETE is asserted refused and fenceless, and the DELETE completes under its recorded (`id`, `generation`) token with no unpointed content resurrected. A **virgin-locator token control** exercises the §3.7 absent-token base case: the FIRST lock grant ever issued for a locator (no recorded acquisition token exists in the brain store) performs its mandatory fence-advance and is asserted ACCEPTED — the advance mints the locator's epoch and its acceptance atomically creates the token record — and the holder's subsequent fenced mutation succeeds; a delayed-first-grant race variant suspends that first holder's advance before application, terminates the session at the deadline, lets a second holder acquire the lock (receiving a greater token) and complete its own advance, then resumes the first holder's advance and asserts it REFUSED by the token comparison — the base case never admits a superseded first grant. A **pre-advance linearization control** exercises the §3.7 legal-outcome boundary: a terminated holder's stale content mutation is scheduled to linearize immediately BEFORE the next holder's fence-advance swap; the control asserts the store serializes the stale mutation first, the advance preserves the post-mutation content presence and bytes and mints its epoch after it, the new holder's first relied-on observation reflects the committed stale effect, and the end state is consistent — equivalent to the stale operation having completed before the previous holder's lock release. The control runs BOTH database-half branches: with the stale mutation a managed deletion's step-1 brain half, its stale step-2 DELETE remains conditioned on the recorded (`id`, `generation`) pair and is asserted to match ZERO rows against a later row incarnation; with the stale mutation an ordinary §3.7 writer's step-1 content commit, the terminated run is asserted to perform NO pointer upsert — the unpointed content is repaired by §4.5 within one cycle, and a retry of the write is asserted to run as a fresh lock acquisition under a new token. An **exceedance control** contrives a cycle that cannot finish within its computed `M` (an injected sequence of deadline-length holders on a reserved locator) and asserts the cycle reports FAILURE — never completion — surfaces the error, withdraws its pending reservations, and persists the carry-over set; the next triggered cycle is asserted to take the carry-over locators into scope first and delete the orphan before reporting completion. A static control asserts the deletion implementation conditions on BOTH the recorded row `id` and `generation` — not on `content_hash` and not on `generation` alone — a second static control asserts the managed-deletion and reconciliation-repair implementations each acquire the §3.7 locator lock before their observation reads (§4.5, §4.7), and a third static control asserts the reconciliation implementation reserves a failed try-acquire's queue place at skip time for a within-cycle blocking acquire — an implementation that defers a skipped locator to a later cycle other than through the §4.5 exceedance transition (which fails the cycle and persists the carry-over set), or that reports cycle completion with a reserved locator unprocessed, FAILS. A fourth static control asserts every lock-holding path (ordinary write §3.7, managed deletion §4.7, reconciliation repair §4.5) is configured with the declared §3.7 hold deadline — an implementation with any unbounded lock-holding path FAILS. A fifth static control asserts the §3.7 grant-order binding: the implementation either cites the pinned advisory-lock mechanism's documented queue-order grant guarantee or routes conflicting locator-lock acquisitions through an application-level FIFO queue — an implementation relying on an undocumented grant order FAILS. A sixth static control asserts the managed-deletion brain step applies its mutation conditionally on the recorded §3.7 fence epoch — an unfenced brain-deletion path FAILS. A seventh static control asserts the §3.7 fence-epoch invariants: every fenced-mutation path mints a fresh never-reused per-locator epoch atomically with its mutation — byte-identical recommits and transitions to absence (the unique tombstone) included — every lock-acquiring path that mutates or relies on brain state (ordinary write §3.7, managed deletion §4.7, reconciliation repair §4.5) performs the fence-advance immediately after acquisition and records the post-advance epoch as its fence value, and no path represents the fence by content bytes, content hash, or bare path absence — an implementation whose fence value can recur, or that observes before advancing, FAILS. The same control asserts the §3.7 acquisition-token invariants: the lock authority mints a fresh, strictly increasing per-locator token at every lock grant; every brain-mutation path — the fence-advance included — presents its holder's token unchanged across every swap-loop retry; the brain store's token comparison and recording are atomic with the mutation they guard; the advance is refused when its token is less than the recorded value; and a refused advance yields no usable fence — an implementation whose advance applies unconditionally, that re-mints or refreshes a token within an acquisition, or whose holder proceeds to any brain observation or mutation after a refused advance, FAILS. 11. **Default-deny and granularity witnesses:** an agent grantee with no active row is refused; with a `granted` row for category A only, category B is refused; with agent X granted, agent Y of the same type is refused for the same (user, category); `connector` and `feature` are exercised via their FK refusal paths, since both registries are empty at version 1 (§5.2); a consent insert naming a nonexistent agent id violates the `agent_id` FK, and the witness pins that FK's referenced table by symbol identity to the `agents` table exported by the shared schema package (§5.2 binding); a platform admin requesting another user's sensitive content through any production read surface is refused — admins are not a grantee type and default-deny applies (the access-time complement of §7.12's write-time refusal); a hierarchy owner or manager grant over the subject confers no consent access (§5.7); with the consent state unreadable, evaluation refuses (§5.6). 12. **Mutation-authority witnesses:** a platform admin attempting to create a grant on another user's data is refused at write time; a hierarchy owner or manager likewise; the data subject succeeds; `actor` equals the subject's principal on every mutation row — there is no non-subject mutation path (§5.4). **Cessation cascade:** deleting an agent's `agents` row removes that agent's consent rows in the same transaction via the FK cascade, the next consent evaluation for that grantee refuses (no row), the linked audit events survive, and no other grantee's rows are touched; a connector-registry amendment migration removing a key cascades identically (§5.4). **Cessation events:** the same deletions each produce one consent-cessation event per removed row in the §1.8 audit event table, carrying the §5.4 fields (cause with parent table and key, actor — the surface-set actor when one initiated the deletion, `system:cascade` otherwise — the grant snapshot including the deleted consent row's `id` and `user_id`, and time); a **two-subject discrimination witness** has two distinct users each hold a `granted` row to ONE agent grantee for the SAME category, deletes that grantee's `agents` row, and asserts exactly two cessation events exist and that each event's snapshot carries a distinct consent-row `id` and the correct `user_id` — every event names exactly which subject's grant of which row ended; an atomicity control rolls the deleting transaction back after the parent DELETE and asserts no cessation event persists (no orphan event without a deletion, no deletion without its events); the events survive subsequent operation; and a subject-cessation (`users` row) deletion produces its events identically (§5.4). 13. **Revocation/re-grant witnesses:** after revocation commits, the next evaluation refuses and the revoked row persists unmutated; a full grant → revoke → re-grant cycle yields two rows (one revoked, one active) and access follows only the active row; a second concurrent grant attempt for the same key violates the §5.1 partial unique index; a direct SQL UPDATE flipping a revoked row back to `granted`, an UPDATE altering any other column of a revoked row, and a standalone DELETE of a row whose subject and grantee both still exist are each rejected by the §5.5 trigger; the complement controls pass: deleting the subject's `users` row cascades the subject's consent rows, and deleting the grantee's referent row cascades that grantee's rows — both permitted by the trigger's absent-referent prong (§5.5). 14. **Self-access witness:** the data subject reads their own content without consent rows (§5.6). 15. **Mode witnesses:** the custody-table schemas are byte-identical under Standalone and Enterprise migrations (§6.1); a fresh Standalone bootstrap records `standalone_layout = 'split'` by default, and an explicit opt-out records `unsplit` with actor and audit linkage (§6.3); conversion from an `unsplit` install refuses the mode flip until partitioning has produced the per-user region (§6.5, with contract 6 §4.2); in an Enterprise fixture with two users, user A's grantee with a grant on user A cannot reach any of user B's content (§6.6); inserting a second `custody_config` row violates the §6.3 singleton constraint. 16. **Ordered-write protocol witnesses:** the §3.7 protocol is fault-injected at its interruption points: killed before the brain commit, content presence, content bytes, and every database row are unchanged — the post-acquisition fence-advance's epoch mint and token recording are the only permitted brain change — and the caller receives no acknowledgment; killed between the brain commit and the pointer upsert, the brain holds unpointed content, no pointer row exists, no read surface serves the content, and the caller received no acknowledgment — then the retried write recommits the same deterministic locator, completes the pointer upsert, and the witness asserts the converged state (one pointer, verifying hash, content served to the subject); killed after the pointer commit but before the acknowledgment reaches the caller, the retry executes under the locator lock, observes the converged state, returns success, and the witness asserts exactly one pointer with a verifying hash and an unchanged answer (idempotent completion, no duplicate content commit that changes the served bytes). **Same-locator interleaving:** two concurrent writers to one (user, category, slot) run to completion; the witness asserts both are serialized by the §3.7 advisory lock (their step-1/step-2 spans do not interleave), the final pointer's hash verifies against the blob committed at the locator, and no `mismatch_at` is stamped by a subsequent read; a control with the lock removed schedules the divergent order (A's pointer upsert after B's content commit) and asserts the witness DETECTS the divergence — proving the lock is what excludes it. **Stale retry:** writer A is interrupted before acknowledgment; writer B then completes a full write to the same locator; A's retry executes under the lock and the witness asserts the end state is consistent (the pointer's hash verifies against the current blob — last completed write wins, no permanent mismatch). **Locator injectivity:** the declared scheme maps every distinct (category_key, slot) pair in the registry to a distinct `brain_ref` (exhaustive at version 1: seven categories x the `answer` slot), and a deliberately colliding scheme control fails the witness. A static control asserts the implementation orders content before pointer: no enumerated sensitive-write route creates a pointer row before its brain commit has been confirmed (§3.7). A second static control asserts every enumerated sensitive-write route applies its step-1 brain commit as a fenced mutation conditioned on the recorded §3.7 fence epoch — recorded at the route's post-acquisition fence-advance, the commit itself minting a fresh epoch and carrying the route's §3.7 acquisition token — an unfenced or token-free brain-commit path FAILS. ## 8. Drafting additions (PRD §12.1 disclosure) The following are proposed drafting additions, visible here for ratification; none is claimed as a PRD mandate, and each is severable: 1. The `feature` grantee type, with a contract-owned registry table that is empty at version 1 (§5.2). 2. Append-only consent history: re-grants insert new rows; revoked rows are retained unmutated, enforced by trigger (§5.5). 3. Cessation by cascade: when a grantee or subject referent row is deleted, consent rows are removed by `ON DELETE CASCADE` in the same transaction; access ends through default deny, and durable history is carried by the §1.8 audit event table. The trigger's DELETE prong permits a child-row DELETE only when the referenced subject or grantee row is absent (§5.4, §5.5). Each cascaded deletion additionally emits a consent-cessation audit event — written by an AFTER DELETE trigger in the same transaction as the referent deletion, carrying cause, actor (surface-set or `system:cascade`), the grant snapshot — which, per revision 6, includes the deleted consent row's `id` and its `user_id`, so events from different subjects' grants to one ceasing grantee remain distinguishable — and time — so the audit trail records when and why each active grant ended (§5.4, §7.12). This replaces the revision-2 system-actor auto-revocation predicate, which is withdrawn. 4. The `custody_config` election record for the Standalone layout, singleton by constraint (§6.3). 5. The domain-separated keyed `content_hash` construction, its key rotation rule, and the `mismatch_at` pointer state (§4.4, §4.6). 6. The bounded dangling-pointer reconciliation policy and the ordered managed-deletion protocol (§4.5, §4.7). 7. The column-type allowlist verification requirement (§7.2). 8. The amendment-only, versioned registry classifier itself, its unknown-key refusal, and both reclassification transition policies (§2.1, §2.3, §2.4). 9. The custody-config write fence and the declared supported-version refusal, including the revision-4 pinning: both fence transactions run at `READ COMMITTED` with the migration's `SELECT ... FOR UPDATE` as the first statement of its transaction, and the supported version is an exported compile-time constant in the custody module of the shared schema package, read under the same lock as the decision it gates (§2.6, §2.7). 10. Subject-only consent mutation authority (§5.4, the single subject predicate). 11. The declared non-sensitive profile table `profile_answers` and its closed schema (§3.5). 12. The typed grantee referent columns, the `custody_connector_registry` and `custody_feature_registry` tables, the audit-table binding by identity (§5.1, §5.2, §1.8), and the agent-grantee binding by symbol identity to the `agents` table exported by the shared schema package (§5.2, §7.11). 13. Shipping all §2–§6 schemas in v1 ahead of any sensitive write surface (§6.4). 14. The committed-tree-only resolution rule: content reads resolve only through the committed tree of the brain repository, and symlink or non-tree/non-blob resolution is refused (§4.3, §7.8). 15. The text user-key binding: every `user_id` column is `text` matching the live `users.id` declaration, values are opaque strings not guaranteed to be UUIDs, and any sibling user-key migration triggers amendment of this contract (§3.6, §7.1). 16. The ordered sensitive-write protocol for ordinary writes: content-first at a deterministic locator, pointer upsert second in its own transaction, acknowledgement only after the pointer commit, idempotent retry convergence, and the pointer-before-content prohibition — with the revision-5 locator scheme (injective over (category, slot), the slot set fixed to `answer` at version 1, stable across releases) and the per-locator advisory lock spanning both steps — held, per revision 6, by EVERY blob/pointer mutator for the locator: ordinary writers, managed deletion, and reconciliation repair (try-acquire with the revision-7 within-cycle blocking retry, so a failed try-acquire defers repair to the end of the cycle, never past it; per revision 8 every hold runs under a declared finite hold deadline with arrival-order grants, so the deferred repair — and with it cycle completion — is wall-clock bounded, not merely eventual; per revision 9 the deadline is enforced by a supervisor independent of the holder, the deferred repair reserves its queue place at the moment its try-acquire fails, the cycle maximum is a computed formula over reservation-point snapshots that is fixed when the last reservation is placed, exceedance is an explicit failed-cycle transition with a durable carry-over set the next cycle processes first — so the one-cycle orphan bound binds every cycle that reports completion — and every brain mutation under a locator lock is fenced by a conditional application; per revision 10 the fence value is an ABA-resistant per-locator fence EPOCH — minted fresh by every mutation, byte-identical recommits and transitions to absence (unique tombstone) included, and advanced by every new holder immediately after acquisition, before it relies on brain state — so a stale tail is refused even when bytes or absence recur and even before the new holder's first mutation; per revision 11 the fence-advance itself is guarded by an acquisition token — minted fresh and strictly increasing by the lock authority at every lock grant, carried unchanged across an acquisition's brain operations and swap-loop retries, compared and recorded by the brain store atomically with every brain mutation, the advance included — so a superseded acquisition's delayed advance is refused and mints nothing, a refused advance yields no usable fence with a mandatory abort-release-surface transition, and a stale operation either serializes wholly before the new holder's advance — the defined legal pre-release-completion-equivalent outcome — or is refused; per revision 12 the token comparison is exact at both ends — strictly less-than refuses, an absent recorded token compares lower than every minted token so a virgin locator's first advance is accepted and creates the record — and the pre-advance-linearization outcome distinguishes the stale database halves: pair-conditioned step 2 for deletion and repair, no pointer upsert ever for a terminated ordinary writer), so live interleavings between any two mutators are excluded by one serialization protocol, and stale external tails are excluded by the fence (§3.7, §4.5, §4.7, §7.16). 17. The compare-and-delete managed-deletion step: deletion step 2 is conditioned on the PAIR of the target row's uuid `id` (the immutable row-incarnation identity) and its monotonic `generation` counter, both recorded in step 1 under the locator lock (revision 6; the revision-4 `content_hash` token was withdrawn as ABA-prone under identical-content rewrites, and the revision-5 bare `generation` token is withdrawn as ABA-prone under delete/reinsert reincarnation, where the counter restarts at 1), so a stale deletion surviving a crash window is never able to unpoint a fresh row — changed, byte-identical, or reincarnated (§4.1, §4.7, §7.10). 18. The sensitive-path write-set closure: a static inventory of every table a sensitive-write route can write, required to equal exactly the pointer table plus the §1.8 audit event table (§7.3). The revision-1 "reporting" rationale for relational background/education storage is withdrawn; the traced rationale is agent configuration (PRD D4). ## Ruling request Ruling requested (one decision): classify `professional-background` and `education` as **non-sensitive** in the version-1 registry (stored relationally, used for agent configuration) — or, as the alternative, classify both **sensitive** (user-brain custody with pointers), accepting that agent-configuration reads then go through pointer indirection and consent evaluation?