WI-3 (#830): compaction observers → revoke + D4 same-PID generation auto-revoke (#842)

docs/architecture/compaction-revocation.md

@@ -0,0 +1,43 @@
+# Compaction observer revocation and runtime generations
+
+WI-3 connects Claude and Pi compaction/session lifecycle events to the existing authenticated lease-broker state machine. It does not add a second lease store or let runtime hooks assert identity. Each observer inherits the broker-minted session, resolves the current private runtime generation, and sends the existing `revoke_lease` action over the authenticated Unix socket.
+
+## Observer matrix
+
+| Runtime          | Lifecycle signal                                                 | Action                                                                                                                       |
+| ---------------- | ---------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------- |
+| Claude / Claudex | `PreCompact`                                                     | Revoke the current lease before compaction. A non-zero hook result blocks the lifecycle transition.                          |
+| Claude / Claudex | `SessionStart` with matcher `compact`                            | Revoke again after compacted context starts.                                                                                 |
+| Claude / Claudex | `SessionStart` with matcher `resume\|clear`                      | Atomically advance the private generation, then revoke the replacement incarnation.                                          |
+| Pi               | `session_before_compact`                                         | Revoke before compaction; return `{ cancel: true }` if revocation cannot be confirmed.                                       |
+| Pi               | `session_compact` then the first `context`                       | Arm and run an independent post-compaction revoke. A failed post observer blocks later tools locally until a retry succeeds. |
+| Pi               | `session_start` with reason `reload`, `new`, `resume`, or `fork` | Atomically advance the private generation, then revoke the replacement incarnation before reuse.                             |
+
+The first observer that reaches the broker deletes pending promotion tokens and makes the lease `UNVERIFIED`. The second compaction observer is deliberate redundancy, not a prerequisite for the first. Claudex receives the same mandatory hooks in its isolated `CLAUDE_CONFIG_DIR`; hook merging preserves unrelated isolated settings and rejects malformed or symlinked settings fail-closed.
+
+## Private generation authority
+
+`launch-runtime.py` still registers before `exec`, preserving the kernel-authenticated PID/starttime anchor. It now also creates `generation-<broker-session>.state` beside the broker socket. The file is owner-only mode `0600` under the broker's mode-`0700` directory. Hook descendants read that file instead of relying only on an immutable inherited environment value.
+
+Generation changes use an exclusive file lock, validate owner/type/mode/size, increment monotonically, truncate and write the complete new value, and `fsync` before contacting the broker. Therefore reload, new-session, resume, and fork events may retain the same PID/starttime while still becoming a new broker incarnation. The higher generation causes the broker to atomically discard prior tokens and lease authority; the replacement generation inherits no VERIFIED lease.
+
+If an observer fires while broker transport is unavailable, `revoke-lease.py` advances the private generation as a local fence before returning non-zero. Every later all-tools gate reads that higher value. When the broker is reachable again, authentication of that value performs the same old-generation revocation before authorization. Pi also keeps a process-local post-compaction/rollover failure latch that blocks tool calls. An unsafe or unreadable generation file itself makes both lifecycle revocation and tool authorization fail closed.
+
+## Threat contract and stopping boundary
+
+### BOUNDED RESIDUAL STALE WINDOW
+
+If **both** pre- and post-compaction observers are missed entirely, no revocation signal exists. During the remaining unexpired lease, **within-TTL consequential actions are allowed**. Their count and timing are **bounded by lease expiry, not by the mutator gate**. WI-3 makes no claim that it bounds mutator actions inside this stale interval. The broker's monotonic lease TTL is capped at 300 seconds; after expiry, the next consequential tool is denied with `LEASE_EXPIRED`.
+
+This is the named D2-v5 T-A residual. It is distinct from an observer that fires but cannot contact the broker: the latter creates a local generation fence and fails closed. It is also distinct from T-C total rot, where the lifecycle observers and the all-tools gate are both absent or replaced. Server-side branch protection, required CI, and independent review remain the irreducible backstop for T-C.
+
+| Condition                                                       | Result                                                                                                                |
+| --------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------- |
+| Either compaction observer succeeds                             | Existing lease and pending promotion tokens are revoked immediately.                                                  |
+| Observer runs but broker confirmation fails                     | Lifecycle transition is denied where supported; local generation fence and runtime latch prevent inherited authority. |
+| Both observers are missed, lease unexpired                      | **ALLOWED** inside the bounded residual stale window. No within-window mutator bound is claimed.                      |
+| Both observers are missed, lease expired                        | **DENIED** by monotonic TTL expiry.                                                                                   |
+| Generation advances on reload/new/resume/fork                   | Prior incarnation revoked; replacement starts `UNVERIFIED`.                                                           |
+| Lifecycle observers and all-tools gate both fail or are removed | T-C total-hook-miss residual; protected-branch controls remain required.                                              |
+
+The permanent T12b/T30 acceptance case prints both required outcomes: dual-hook miss within TTL is **ALLOWED**, and the same lease after TTL is **DENIED**. Separate real-socket tests prove each Claude observer and same-PID generation rollover; Pi lifecycle tests exercise pre/post observers, all four replacement reasons, and local failure closure.

docs/architecture/lease-broker-protocol.md

@@ -14,10 +14,10 @@ Each connection carries exactly one UTF-8 JSON object followed by one newline, c
 - `consume_token`: authenticated identity plus `token`.
 - `begin_verification`: authenticated identity, runtime (`claude` or `pi`), cycle `binding`, and a TTL no greater than 300 seconds. The broker revokes existing authority first, enters `PENDING_VERIFICATION`, and returns a single-use promotion token.
 - `promote_lease`: authenticated identity plus the exact pending promotion token. The broker commits token consumption before making `VERIFIED` visible.
-- `revoke_lease`: authenticated observer signal; deletes pending tokens and makes the session `UNVERIFIED` immediately.
+- `revoke_lease`: authenticated observer signal; deletes pending tokens and makes the session `UNVERIFIED` immediately. WI-3 Claude/Pi hooks send this existing action; `runtime` and bounded `reason` fields are diagnostic input only and never identity authority.
 - `authorize_tool`: authenticated identity, runtime, and exact runtime-reported tool name. The broker returns an explicit allow/deny decision from the whole-class policy and current lease.
 
-A higher generation for the same anchor atomically replaces the stored incarnation and deletes all prior tokens and lease authority for that session. A lower generation is stale. Tokens are 256-bit values from the operating-system cryptographic RNG and are single use. At most 256 pending tokens may be persisted; another mint fails with `TOKEN_CAPACITY` before mutation. Successful consumption deletes the token, while a replay still fails with `TOKEN_REPLAY`. Live v1 token records retain the existing `consumed: false` schema.
+A higher generation for the same anchor atomically replaces the stored incarnation and deletes all prior tokens and lease authority for that session. A lower generation is stale. Runtime descendants resolve the current generation from an owner-only, locked generation file created by the register-before-exec launcher; reload/new/resume/fork observers advance and `fsync` it before broker revocation. This supports generation replacement even when PID/starttime do not change. Tokens are 256-bit values from the operating-system cryptographic RNG and are single use. At most 256 pending tokens may be persisted; another mint fails with `TOKEN_CAPACITY` before mutation. Successful consumption deletes the token, while a replay still fails with `TOKEN_REPLAY`. Live v1 token records retain the existing `consumed: false` schema.
 
 VERIFIED leases are volatile and monotonic-time bounded: broker restart, generation change, explicit observer revocation, or expiry returns the session to `UNVERIFIED`. `begin_verification` always revokes before minting a new prerequisite. `promote_lease` is valid only from the matching pending cycle; persistence failure rolls token and lease state back, while post-rename durability uncertainty terminates the broker. The WI-1 token is the atomic promotion prerequisite substrate. A later receipt implementation must satisfy that prerequisite but cannot replace the mechanical mutator gate as safety authority.
 

docs/architecture/lease-broker-security.md

@@ -2,13 +2,25 @@
 
 - Trusted identity comes only from Linux `SO_PEERCRED` plus `/proc` starttime, never request identity fields.
 - Descendant authorization is anchored to `(pid,starttime)` and uses a complete second starttime pass to fail closed on disappearance or PID-reuse races.
-- Runtime generations are monotonic per anchor; a bump revokes prior-incarnation tokens before persistence commits.
+- Runtime generations are monotonic per anchor; a bump revokes prior-incarnation tokens before persistence commits. WI-3 stores the live generation in an owner-only locked file so same-PID Pi reload/new/resume/fork and Claude resume/clear transitions cannot inherit a VERIFIED lease.
 - Session IDs and cycle tokens use the OS cryptographic RNG. `Math.random` and model output are not token sources.
 - Framing and persistence failures fail closed. Sensitive tokens are not logged.
 - Built-in `0700`/`0600` filesystem modes provide same-principal hardening only, not socket authenticity against the same UID. WI-1 provides no distinct-principal isolation. That stronger deployment requires an external protected proxy, ACL, or service boundary, and the boundary must preserve authenticated client identity for the broker's `SO_PEERCRED` and ancestry authorization rather than substituting a shared proxy identity.
 - WI-2 whole-class authorization denies every consequential, unknown, and custom tool while UNVERIFIED; it does not inspect shell strings or trust wrapper selection. First-class Claude/Pi, both Claudex dispatch modes, PRDY, QA remediation, coord, orchestrator, and fleet starts converge on broker register-before-exec; Claudex additionally installs the mandatory all-tools hook inside its preserved isolated config and fails closed on unsafe settings.
 - The permanent `check-runtime-launches.py` suite/CI guard scans production source for direct literal, absolute-path, process-API, command-array, and dynamic Claude/Pi launches. It has no bypass allowlist: an unrecognized launch form fails CI until routed through the common boundary.
 - WI-2 promotion consumes a WI-1 cycle token before VERIFIED becomes visible. Observer revocation, runtime-generation replacement, broker restart, and monotonic TTL expiry remove authority.
-- Receipt observation, payload construction, compaction observers, and constrained recovery implementation remain later surfaces. A receipt can become a promotion prerequisite but is never the safety mechanism.
+- WI-3 wires redundant Claude `PreCompact`/`SessionStart(compact)` and Pi `session_before_compact`/post-`session_compact` `context` observers to that same revoke action. If broker confirmation fails after an observer fires, the revoker advances the private generation as a local fence; subsequent authorization revokes the stale broker incarnation before any consequential allow.
+- Dual observer absence while a lease remains live is the named **bounded residual stale window**: consequential tools remain allowed until monotonic expiry, with no claimed within-window action bound. After expiry they are denied. Total observer-plus-gate absence remains T-C.
+- Receipt observation, payload construction, and constrained recovery implementation remain later surfaces. A receipt can become a promotion prerequisite but is never the safety mechanism.
+
+## Named residual: promote-lease-lost-ACK (WI-3 D2-v5)
+
+A valid `promote_lease` can leave a session `VERIFIED` in the broker while the client never learns of it. This is a named, bounded D2-v5 T-A residual — an **authority-observability divergence, not an authority divergence, not an ALLOW-risk, and not a retry double-apply**. It is disclosed here, not laundered.
+
+**Window — where it can occur.** The broker commits token consumption and durable `VERIFIED` state _before_ the success reply becomes visible (see the promotion order in `lease-broker-protocol.md`). The residual is confined to the interval after that commit+fsync when the broker→client reply or peer-ACK is lost — for example an extreme-contention send failure or peer disconnect after `handle()` has already mutated and persisted state (the #838 fail-closed transport path). The lease mutation is already durable broker-side; only the acknowledgement to the client is lost. No uncommitted or partially-applied state is involved: the commit either happened (and is authoritative) or it did not (and no lease exists).
+
+**Fail-safe direction — the client can only under-claim.** Broker intent is the ceiling; client authority is always ≤ broker intent, never more. Client-side authority-belief is granted only by a _received_ acknowledgement; a lost acknowledgement conveys nothing, so the client cannot conclude "verified" and continues to treat itself as `UNVERIFIED` (it re-verifies or recovers). If the client retries `promote_lease` with the same token, the token is already consumed and the broker rejects the retry (`PROMOTION_TOKEN_MISMATCH` / `INVALID_LEASE_TRANSITION`); there is no double-apply. The committed `VERIFIED` state the broker holds is authority the lease _legitimately earned_ from a real promotion — the broker authorizing consequential tools under it is correct, not inflation. Divergence is therefore strictly toward _less_ client authority than the broker granted; it never produces authority the broker did not grant.
+
+**Bound — TTL plus the observer/gen-bump revoke backstop, self-healing.** The orphaned `VERIFIED` lease is indistinguishable to the broker from any other legitimately verified lease, so the identical D2-v5 revocation backstops dispose of it: any compaction observer (`PreCompact` / `SessionStart(compact)` for Claude; `session_before_compact` / post-`session_compact` `context` for Pi), any same-PID runtime-generation bump (reload/new/resume/fork), broker restart, or monotonic-time expiry returns the session to `UNVERIFIED`. Monotonic TTL expiry (capped at 300 seconds) is **unconditional** — it requires no observer at all — so the maximum exposure of the orphaned lease is one TTL, ≤ 300 s, after which the next consequential tool is denied with `LEASE_EXPIRED`. Any observer that fires shortens the window further. The residual self-heals: "≥1 observer fires OR expiry ⇒ revoke" catches the lost-ACK lease on the same terms as every other stale lease. As with the dual-observer-miss stale window, WI-3 makes no claim that the mutator gate bounds actions inside the residual interval; the interval is bounded by TTL and the revoke backstop, and the server-side branch-protection / required-CI / independent-review line remains the irreducible backstop for protected-repository mutations.
 
 Coordinator security review must rerun the real socket/peercred and mutator-gate acceptance suites on an unrestricted Linux runner and obtain the mandated independent Opus-SECREV review before integration.

docs/architecture/mutator-class-gate.md

@@ -21,13 +21,13 @@ The authenticated broker is the sole lease writer:
 3. Token consumption commits before the volatile lease becomes VERIFIED. Promotion is last and cannot be reached directly from UNVERIFIED.
 4. `revoke_lease`, a runtime-generation increase, broker restart, or monotonic expiry removes mutator authority.
 
-The initial TTL is capped at the ratified 300-second maximum. A caller may request a shorter positive TTL but cannot lengthen the maximum. Dual compaction-hook miss within an unexpired lease remains the ratified bounded T-A residual; once either observer revokes or TTL expires, the next consequential tool is denied.
+The initial TTL is capped at the ratified 300-second maximum. A caller may request a shorter positive TTL but cannot lengthen the maximum. WI-3 installs the [compaction observer and generation lifecycle](compaction-revocation.md). Dual compaction-hook miss within an unexpired lease remains the ratified bounded T-A residual: consequential tools are allowed until expiry, with no claimed within-window action bound; once either observer revokes or TTL expires, the next consequential tool is denied.
 
 A receipt is only a future promotion prerequisite. It is not an obedience, residency, or safety proof and never replaces this mechanical gate.
 
 ## Runtime adapters
 
-`launch-runtime.py` registers itself with the broker and then `exec`s Claude or Pi so PID/starttime remain the authenticated parent anchor. It exports only the broker-minted session ID and current generation to descendants.
+`launch-runtime.py` registers itself with the broker and then `exec`s Claude or Pi so PID/starttime remain the authenticated parent anchor. It exports the broker-minted session ID and an owner-only generation-file reference to descendants; lifecycle hooks advance that file for same-PID replacement generations.
 
 - Claude installs `mutator-gate.py` as an all-tools (`.*`) `PreToolUse` hook.
 - `mosaic claudex` and `mosaic yolo claudex` preserve their isolated `CLAUDE_CONFIG_DIR`, merge the mandatory hook into that isolated `settings.json`, and use the same register-before-exec launcher. Malformed or symlinked isolated settings deny launch.