Correct Windows shutdown observation and record compiler interruption

.gitattributes

@@ -78,3 +78,5 @@ provenance/video-tool/** -text whitespace=cr-at-eol
 provenance/video-tool/predecessor/src/adaptive_inference/desktop/hermes/report_contract.py whitespace=cr-at-eol,-blank-at-eof
 # Exact historical CI fixture bytes include their original CRLF line endings.
 provenance/ci-fixtures/** -text whitespace=cr-at-eol
+# Preserve the observer version used by the retained physical failure.
+provenance/video-coexistence/observer-shutdown/** -text whitespace=cr-at-eol

CHANGELOG.md

@@ -4,18 +4,26 @@
 
 These changes belong to the development source. The published `0.4.0a1` tag and downloadable packages retain their original capabilities and validation scope.
 
+- Validate **active compiler interruption** through one installed Hermes SDK worker: durably retain interrupted feedback, hold the next provider request while foreground activity remains busy, then complete a distinct compiler call within the original task budget. Independent closed review passed 61 checks. The compiler/SDK were real; provider replies and foreground/resource signals were scripted. Actual IQ3_S/physical active-tool qualification remains open. See the [scoped evidence](provenance/compiler/installed-active-tool-interruption-2026-11-09.json).
+- Correct the GET-only Windows shutdown observer's connection bound after a known instance and terminal marker: allow up to five seconds, clamped to its remaining outer allowance, while preserving refusal-only closure acceptance. Ten component checks and two affected-test follow-ups passed; stalled responses still fail. Earlier physical observer failures remain failed. See the [observer record and retained predecessor](provenance/video-coexistence/observer-shutdown/record.json).
 - Add **Open saved task** after a safe Stop, including tasks created later through the running controller. Accepted installed IQ3_S trials preserve the original task, confirmed reads, conversation, deadline and consumed work across application close/reopen in the same Windows boot. A completed task can be viewed without starting another worker or model, or writing its result again. See [saved-task reopening](docs/saved-task-reopening.md).
 - Add a bounded **C++17 compile-and-test recipe** that an opted-in Hermes task can request through the existing resource handoff. Reuse CMake/CTest, retained process ownership and durable tool feedback. An accepted IQ3_S foreground trial handed resources to the compiler and returned its checked result to the same task. See [the compiler recipe and evidence](docs/compiler-tool-recipe.md).
 - Add a bounded **GPU video recipe** using pinned FFmpeg/FFprobe and NVENC. An accepted IQ3_S foreground trial parked inference, encoded and checked one silent 90-frame test clip, restored the model and completed the task from saved tool feedback. This is a fixed test-clip capability; H3, audio synchronization and the full autonomous media workflow remain unqualified. See [the video recipe and evidence](docs/video-tool-recipe.md).
 - Recover missing tool feedback through the existing explicit reconciliation path when the completed result is known. Extend this to a pending journal entry when the live tool host retains the exact terminal result. Installed Hermes SDK trials use one actual video operation with scripted provider replies and resource observations; they preserve the task and deadline without replaying the operation. Unknown outcomes remain unresolved. See [completed-result recovery](docs/video-tool-recipe.md#completed-result-feedback-repair) and the [pending-effect receipt](provenance/pending-tool-recovery/installed-worker-loss-2026-11-02.json).
-- Add **automatic desktop worker replacement** for that narrowly defined pending-result boundary. In the accepted real IQ3_S trial, foreground activity parked inference, one video operation finished, and the first worker exited before recording completion. The live host reconciled the saved result and launched one replacement automatically; the same task completed within its original deadline after two total model requests. Independent closed review passed 49 checks. The separate shutdown observer failed and remains an open diagnostic; this is not active-tool interruption, whole-host loss or reboot recovery. See [automatic recovery](docs/video-tool-recipe.md#automatic-desktop-recovery) and its [physical receipt](provenance/automatic-tool-recovery/physical-foreground-recovery-2026-11-08.json).
+- Add **automatic desktop worker replacement** for that narrowly defined pending-result boundary. In the accepted real IQ3_S trial, foreground activity parked inference, one video operation finished, and the first worker exited before recording completion. The live host reconciled the saved result and launched one replacement automatically; the same task completed within its original deadline after two total model requests. Independent closed review passed 49 checks. The separate shutdown observer remains a failed result; the later component correction above does not requalify it. This trial does not establish active-tool interruption, whole-host loss or reboot recovery. See [automatic recovery](docs/video-tool-recipe.md#automatic-desktop-recovery) and its [physical receipt](provenance/automatic-tool-recovery/physical-foreground-recovery-2026-11-08.json).
 - Add the **durable reboot-budget foundation**: a fixed expiry and saved remaining-time cap prevent a restart from granting a new execution allowance. Source, package, installation and scripted SDK checks passed. The first actual Windows restart preserved saved work but the original allowance expired before continuation; execution was correctly refused. Successful computer-reboot continuation remains open. See [budget behavior](docs/reboot-budget.md) and the [retained expired attempt](provenance/installed-reopen/reboot-budget/physical-restart-expired-2026-11-02.json).
 - Add browser-video observation and comparison tooling, retain unsuccessful measurement attempts and correct continuation-test cleanup. The attempted comparison did not establish desktop/video performance because its observation conditions differed; no broad coexistence or speed claim follows from it.
 - Make evidence-based agent engineering an explicit goal requirement: bounded changes, meaningful acceptance, independent review, reproducible handoffs and a protected main branch. See the [collaboration practices](docs/multi-agent-collaboration.md#engineering-practices).
 - Correct two Windows CI fixture timing assumptions after retaining their original failures. Use a logical observation clock for an ordinary continuation test and a 30-second whole-fixture join guard for retained control work. Twelve targeted tests passed, including stale-observation refusals; production freshness and task deadlines are unchanged. Connect their exact retained predecessors to the existing historical-source reader after its separate hosted failure. The broader local reader sweep timed out after 118 passing checks and remains incomplete; fresh hosted validation is required. See the [fixture evidence](provenance/ci-fixtures/2026-11-09/record.json).
 
 The automatic-recovery source passed the complete **1,595-test local core suite** and all **16 additional workflow commands**, with the earlier failed test-double attempt retained. Source checks, installed SDK fixtures and physical IQ3_S evidence remain separately identified in the [evidence ledger](docs/evidence.md) and [local validation record](provenance/automatic-tool-recovery/local-validation-2026-11-08.json). These results do not complete M3, M4, M5, daily-use qualification or the full Adaptive-Hermes-Media-Autonomous Test.
 
+The corrected CI-fixture/roadmap successor subsequently passed all **14 original
+hosted jobs**, including **12 full-core and 12 distribution checks**, and merged
+as [PR #40](https://github.com/midhatn/adaptive-inference/pull/40). Its
+[record](provenance/ci-fixtures/2026-11-09/record.json) preserves the successful
+successor and the earlier failed or incomplete attempts separately.
+
 ## 0.4.0a1 — supported Windows reference alpha (2026-11-01)
 
 This prerelease packages the supported Windows/IQ3_S/Hermes reference, ordinary

HANDOFF.md

@@ -63,7 +63,24 @@ passing across direct/aliased callers and historical-source assertions.
 Use the exact current PR checks before integrating; historical or local passes
 do not override a failed required check. Main branch protection is enabled.
 
-**Latest accepted M4 increment:** automatic desktop recovery of a pending
+[PR #40](https://github.com/midhatn/adaptive-inference/pull/40) has now merged
+at `d33c979e1047f9c125ab94788fcbb2b3d8ef6c12` with its exact reviewed tree.
+All 14 original hosted jobs passed, including 12 full-core and 12 distribution
+checks; the [canonical record](provenance/ci-fixtures/2026-11-09/record.json)
+retains those outcomes without changing earlier failures or qualification scope.
+
+**Latest bounded M4 gate — 9 November 2026:** the
+[installed active-compiler interruption case](provenance/compiler/installed-active-tool-interruption-2026-11-09.json)
+passed 61 independent closed-evidence checks. One actual Hermes SDK worker
+interrupted a running CTest operation, durably retained its feedback and completed
+a distinct successful compiler call: three provider requests, two effects and the
+same original task/deadline. Provider replies, foreground/resource observations
+and inference-owner actions were scripted. The
+[observer shutdown correction](provenance/video-coexistence/observer-shutdown/record.json)
+has source/component and owned-loopback evidence; it does not relabel earlier
+physical observer failures.
+
+**Preceding accepted M4 increment:** automatic desktop recovery of a pending
 compiler/video operation after worker loss. `DesktopComposition` now detects
 a non-zero worker exit with one pending effect whose exact terminal result is
 held by the live tool host, exports `CompilerHost.terminal_receipt`, runs the
@@ -79,10 +96,10 @@ separate observer failed during shutdown; the accepted task/cleanup audit does
 not erase that failure. Recovery requires the live host's exact completed result
 and does not qualify interruption of an active tool or loss of that host.
 
-**Next bounded work:** diagnose the retained observer shutdown failure before
-relying on it for a fresh physical acceptance, then follow M4's active-tool
-interruption gate. Reuse existing cancellation and ownership mechanisms and
-research any missing behavior before implementation. Whole-host loss, successful
+**Next bounded work:** qualify active-tool interruption with actual IQ3_S and
+physical foreground activity using the reviewed observer correction. Reuse
+existing cancellation and ownership mechanisms and research any missing behavior
+before implementation. Worker loss during an active tool, whole-host loss, successful
 OS-reboot continuation and broad desktop/media qualification remain open under
 the roadmap. Preserve unchanged IQ3_S quality and the closed accepted/failed
 cases; a new experiment needs a fresh case, not a renewed historical budget.
@@ -359,12 +376,12 @@ daily-use qualification remain open.
 | Integrated main starting point | Commit `f362ed0ed8c0f9f6552bd71c9c3ba9e8a847aa3a`, including PRs #34–#39, has the exact reviewed `08053a0` tree. Verify current `main`; this is the 9 November anchor, not a claim that it will remain HEAD. |
 | Released reference and active work | [v0.4.0a1](https://github.com/midhatn/adaptive-inference/releases/tag/v0.4.0a1) is published at tested source `57f673f8e934e275bc2bd51e85d434dc7025bc9a`; PR #28's merged tree matches it exactly. PR #29's measurement preparation merged after its exact-head checks. The first adaptive diagnostic and following control are closed; their scoped results and unmet comparison gates are recorded in the [evidence ledger](docs/evidence.md#first-adaptive-video-diagnostic). |
 | Installed source and evidence | The latest automatic-recovery [physical receipt](provenance/automatic-tool-recovery/physical-foreground-recovery-2026-11-08.json) binds its actual installation. Earlier later-task M5 used `b35b631aca9ec37da9955903b87463201a6b3630`; initial-task M5 used `6ac601546daf32d96e38b6689a5818e4cc8dafa8`; M3 video used `7306a53085ddc31d53d2ba50e91975c80cfc0014`. All preserved IQ3_S and quality. A source merge does not update those installations. |
-| Current work | Compiler/video handoffs, known-result reconciliation and automatic desktop worker recovery are integrated. The latest physical trial passed 49 closed-audit checks at the completed-tool/surviving-host boundary. Observer shutdown remains unresolved; active-tool interruption and whole-host loss remain open. PR #34's budget source is integrated, but its expired physical restart attempt does not qualify successful reboot continuation. All consumed trials stay closed. |
+| Current work | Compiler/video handoffs, known-result reconciliation and automatic desktop worker recovery are integrated. The latest physical trial passed 49 closed-audit checks at the completed-tool/surviving-host boundary. Active compiler interruption now passes the scripted-provider installed SDK gate; the observer has a component correction. Their fresh physical gate and loss during an active tool or of the whole host remain open. PR #34's expired physical restart attempt does not qualify successful reboot continuation. All consumed trials stay closed. |
 | Preserved M5 same-boot result | The installed same-boot application restart preserved one input read, the task/native session and original deadline; three calls total completed the correct 23-unit answer with one physical write. A third launch viewed the result without additional calls or changes to the eight canonical files. All three app parents joined normally; bounded observer shutdown errors remain retained. See the source-bound M5 record above. |
 | Preserved M2 recovery result | Case17 passed both original audits: 99 primary recovery/handover checks plus the foreground/lineage supplement. Its correct inventory result survived safe expiry and one Continue; a separate welcome task completed before one normal Stop. Six calls produced two physical writes. Welcome has format-only acceptance. Full M2 is broader than that run. |
 | Preserved limits and failures | Original quality, 900-second execution budgets, 2,700-second outer bound and task accounting remain unchanged. Cases14/15/16 keep their original failed outcomes. Case17 retains the rejected file-picker input and 30-second operating-target misses; neither repeated a task control nor violated the strict foreground gate. |
 | Foreground evidence | Case17 retained 25 real document actions, 24 wholly preseal, maximum adjacent interval 54.680872 seconds and preseal tail 37.453106 seconds. The original supplement passed onset, continuous hold and saved-read lineage; the postseal action is excluded from coverage. This is not a desktop-performance or GPU-preemption measurement. |
-| Next bounded work | Diagnose the retained observer shutdown failure first, then qualify active-tool interruption using the existing compiler/video journal and owner, preserving completed effects and feedback. Whole-host loss remains separate. A future M5 attempt needs fresh identity, prompt postboot return and unchanged allowance. M3's bounded diagnostic remains video-surface attribution without inference. Follow PLAN.md before starting any fresh case. |
+| Next bounded work | Qualify active compiler interruption with actual IQ3_S and physical foreground activity, using the reviewed observer and existing journal/owner while preserving task budget and feedback. Loss during an active tool and whole-host loss remain separate. A future M5 attempt needs fresh identity, prompt postboot return and unchanged allowance. M3's bounded diagnostic remains video-surface attribution without inference. Follow PLAN.md before starting any fresh case. |
 | Latest release validation | Corrected source `57f673f` passed all 14 original hosted jobs, final archive/runtime verification and merge-tree verification. All five uploaded assets and the tag were verified at publication. The 48-check watchdog and 21-check refusal corrections retain their own local scope; earlier failed runs remain failed. |
 | M3 publication | [PR #29](https://github.com/midhatn/adaptive-inference/pull/29) merged at `f6c0da1` with the tested `3136e35` tree. Its original 14 head jobs and seven post-merge jobs passed, including portable archive checks. Earlier archive/fixture failures remain retained; these preparation checks do not requalify the installed runtime. |
 | Diagnostic publication and test correction | [PR #30](https://github.com/midhatn/adaptive-inference/pull/30) merged at `c348398` with the tested `72ac717` tree after 14 head jobs and seven post-merge jobs passed. Its original `fd4d483` pair retains 13 passes and one Windows SQLite concurrency-fixture failure. The [bounded correction](docs/evidence.md#sqlite-concurrency-fixture-ownership) keeps test threads in a retained child and joins before cleanup; 18 focused checks passed. Runtime, models and physical records are unchanged. |