# Core Architecture v2 — Final Closure WorkGraph (AEH-V2-0130 / AEH-V2-0135)

**Slice lead:** fresh final-closure slice lead (opencode-go/deepseek-v4.1-flash), 2026-09-30.
**Authority:** AEH self-development — shell, TypeScript and tests are the authority. No AEH operation is launched against this checkout. No commit/push/stash/reset/clean/revert is performed by this lead; the Campaign Director owns checkpointing after a fresh independent `SLICE_ACCEPTED`.
**Scope:** product repair of AEH-V2-0130/0135 (durable operation→resource ownership + terminal/recovery reconciliation) and its required re-verification. Not S15; unrelated accepted work is not reopened.

## Frozen facts at slice start

- Branch `core-architecture-v2`, HEAD `1d1dd973ff585c8a886f4af3a13a9343d6b1d618` (S1–S14 accepted; S13/S14 checkpoints pushed). `git status` clean.
- TARGET `docs/CORE_ARCHITECTURE_V2.md` sha256 `4954b62bfb3a820fc6e5d0369b2d37bc35a94988d19e485f00d30675aafa555a` (verified at start) — byte-identical, must stay untouched.
- S14 acceptance binding: composite buildDigest `4c05da6ee7fe249bcad4adb730f2c08bb34867910eb512171bccb40b63203284`; report sha256 `75b18b6e…`; manifest sha256 `d8c4a69d…`; `review-s14.json` sha256 `f258c510092b4f5f0c213bf4b514370fb6261dc5305ac9216f97fb5d8037a565`; stability summary sha256 `090dace9…`.
- Open carries: `AEH-V2-0130` (P1, resource lifecycle/reconciliation; product-side durable ownership is the missing half), `AEH-V2-0135` (S14 confirmation of the same defect), non-blocking `0132/0133/0134` + `paseoSdkResolve` environment drift + S14-F4 test hygiene.
- Live residue at closure start (classified, not killed without durable ownership proof; see W0 census): ~10 AEH fixture/child processes, ~24 `opencode serve` processes, 155 `/tmp/aeh-*` dirs (52 `aeh-composite-test-*` test fixtures), 3 accounted S13-era staging roots (`/tmp/aeh-s13-gov-6KeYap`, `/tmp/aeh-s13-gov-Zo72ph`, `/tmp/aeh-serena-pools`).

## Frozen TARGET quote — section 16 "Self-hosting gate" (binding fragment)

> The final disposable self-modification campaign proves that the source checkout remains untouched, the packed candidate controls only the disposable fixture, all candidate and repair mutations stay inside it, side effects reconcile, and **cleanup succeeds**.

Binding interpretation for this closure: "cleanup succeeds" is product-owned durable operation→resource ownership, terminal/recovery reconciliation, safe release/archive/stop, verification receipt, and stable baseline return — not campaign-side PID/glob cleanup. Section 15 certification lanes, the S13 resource amendment (2026-09-29) and the Round-9/Round-10 cleanup-mistake class remain binding.

## Objective (canonical)

Implement durable product-side operation/resource ownership and terminal/recovery reconciliation so terminal operations return their owned resources to baseline without campaign-side cleanup, then re-verify: focused regressions A–J, real stability (≥2 iterations), disposable self-modification, affected lifecycle rows, S14 composite regeneration, docs reconciliation and full validation.

## WorkUnits (bounded; disposition updated as work completes)

| ID | WorkUnit | Kind | Classification | Disposition |
| --- | --- | --- | --- | --- |
| C-W0 | Reconstruction, frozen TARGET/environment facts, residue census, this WorkGraph (persisted before implementation) | documentation + verification | DETERMINISTIC | DONE — census `docs/evidence/closure/residue-census.json` sha256 `c352ae7eb9b99c86c4dd9f536a9345b3514accbbc321c2d466cf3b6129cadd49`: 74 interesting processes (24 `opencode serve`, 14 serena, 12 AEH-named), 154 `/tmp/aeh-*` dirs (52 `aeh-composite-test-*`), 3 S13-era staging roots present. Nothing killed/removed. |
| C-W1 | Product resource ownership: durable per-operation resource registry/inventory (`src/runtime/operationResources.ts`), ownership proof (operation/candidate/controller epoch + exact identity), classification LIVE_OWNED / TERMINAL_ORPHAN / UNKNOWN_OR_UNOWNED, safe release/archive/stop, idempotent reconciliation, durable verification receipt | implementation | DETERMINISTIC | DONE — `src/runtime/operationResources.ts` (registry `.harness/operations/<id>/resources.json`, receipt `resource-reconciliation.json` (with `cleanupComplete` and per-resource dispositions); workspace snapshot before archive; label-bound `aeh.operation` agent discovery; idempotent already-reconciled; failed releases stay visible). |
| C-W2 | Terminal + recovery integration: terminalization reconciliation for SUCCEEDED/FAILED (CANCELLED keeps its fence), recovery sweep on restart/terminal-entry, resource registration at workspace/participant/lease/staging creation points | implementation | DETERMINISTIC | DONE — `terminalizeOperation` reconciles after the terminal transition; `executeOperation` terminal entry and `startDetachedOperation` startup run the deterministic sweep; workspace resources registered in `ensureOperationWorkspace` with `workspaceDisposition`; DIRECT staging roots registered/removed in `candidates/direct.ts`. |
| C-W3 | Policy-driven backpressure: generic per-operation resource/provider-session ceilings with typed wait/exhaustion disposition (no machine-specific constants) | implementation | DETERMINISTIC | DONE — `RESOURCE_CEILING_EXHAUSTED` typed `AehError`; generic policy `orchestration.operations.resources.{maxOwnedResourcesPerOperation,maxConcurrentProviderSessionsPerOperation}` (defaults 64/16) enforced at registration and at provider-lease acquire; model output cannot widen policy. |
| C-W4 | Focused regressions A–J (terminal→reconciled; live→preserved; unknown→preserved; crash/restart→recovery; idempotent; other-operation→rejected; owner binding; durable receipt; model cannot widen ceiling; repeated terminals no accumulation) | verification | DETERMINISTIC | DONE — `tests/operationResourceReconciliation.test.ts` 12/12 passed (A–J plus failed-release visibility and in-line release idempotency); adjacent controller/runtime/direct suites 103/103 passed; typecheck exit 0. |
| C-W5 | REAL bounded resource-stability campaign (≥2 iterations, product-owned reconciliation as the PASS mechanism, campaign-side PID/glob cleanup disabled/reported-not-used) | verification | HYBRID | DONE — `closure-resource-6` PASS (`docs/evidence/closure/resource-stability-product.json` sha256 `94ed891de6e1d72b4039f7a733b55e6625a94a48df61fd5f0288147c5a157f14`): two real iterations terminal SUCCEEDED, product receipts `cleanupComplete`, census 0/0/0, `campaignSideCleanup: NONE`, `CLOSED_BY_PRODUCT`. The retained superseded failure summaries (run-4/run-5 files) plus the run-1..3 digest/history entries in the step log record each real product ownership gap exposed and closed (monitor process; unlabeled preflight assessor + workspace; inspect-resolved agent workspace). |
| C-W6 | REAL disposable self-modification campaign rerun on the frozen closure candidate + envelope regeneration | verification | HYBRID | DONE — `closure-disposable-1` lane PASS / terminal SUCCEEDED / oracle ACCEPTED / rows `change`+`direct-change` PASS, checkout byte-identical, containment disposable, product receipt 13 dispositions (11 reconciled + 2 inline, 0 failed); envelope `docs/evidence/closure/disposable-self-modification-campaign.json` sha256 `17e9b926cc17aaaafa34efa78222bcb1d6c1119f77b0e0375f398160d0f6e19a`. |
| C-W7 | Affected lifecycle row rerun: real Paseo cancellation/recovery lane on the closure candidate (terminal/recovery code changed) | verification | HYBRID | DONE — `closure-lifecycle-1` cancel+recovery 2/2 PASS, checkout untouched, candidate `d266692d…`; closure artifact `docs/evidence/closure/paseo-lifecycle-cancel-recovery.json` sha256 `ead63493846fb3bb15b1ea7bba083a2486d89b642c62c187407d88ae22c73438`. |
| C-W8 | Staleness: exact source-impact analysis (CURRENT vs STALE_REQUIRES_RERUN) with import/path/behavior reasoning | verification + documentation | DETERMINISTIC | DONE — `docs/evidence/closure/staleness-analysis.json` (16 source-delta entries with executed-path impact; 22-row matrix dispositions: CURRENT for the re-executed rows, `STALE_REQUIRES_RERUN` with bounded-scope reasoning for the remaining accepted rows, `NOT_APPLICABLE` control-center). |
| C-W9 | S14 composite regeneration (manifest + report + digest), affected adversarial/fail-closed checks rerun, resource-lifecycle item backed by PRODUCT reconciliation | verification | DETERMINISTIC | DONE — adversarial 11/11 + 18/18 with 0 failed checks (`docs/evidence/closure/adversarial-negative-checks.json` `c100197d…`); composite `ready: true`, 21/21 items, 47 artifacts, 0 failed, policy digest `1be97df3f7736b76dd048648555874b43e7df5cffc5d1ba98b6a0b555085aef3`, manifest `ff1b5b11…`, report `9b9e2134…`, byte-reproducible; resource-lifecycle item backed by product-owned reconciliation. |
| C-W10 | Docs reconciliation (STATUS/CONFORMANCE/LEDGER/S14 WorkGraph CURRENT sections; historical rounds preserved) | documentation | DETERMINISTIC | DONE — closure final sections added to STATUS, CONFORMANCE, S14 WorkGraph; LEDGER 0130/0135 closed with product evidence; historical round records preserved; 0132/0133/0134, `paseoSdkResolve` drift and S14-F4 left explicitly open. |
| C-W11 | Validation: focused closure regressions, `npm run typecheck`, `npm run build`, `git diff --check`, canonical non-system suite; every failure classified | verification | DETERMINISTIC | DONE — focused closure set 43 suites / 201 tests passed (`docs/evidence/closure/focused-regressions.json` `0d670ab4…`); canonical non-system suite 205 files / 1,391 tests: 1,380 passed / 7 skipped / 4 failed, classified `PRE_EXISTING_REPRODUCIBLE` (paseoSdkResolve 3) and `FLAKE_CONFIRMED` (buildHygiene 1; 6/6 in isolation); typecheck exit 0; build reproduced `d266692d…`; `git diff --check` clean; TARGET byte-identical; `docs/evidence/closure/validation.json` `572c2d1c…`. |

## Architecture (frozen design for this closure)

1. **Durable ownership record.** Every resource AEH can own through a governed operation is registered with exact durable identity: `paseo-workspace` (workspaceId + root), `paseo-agent`/`provider-session` (agentId/sessionId + owner participant/generation), `managed-process` (registered handle pid + process group), `staging-root` (exact absolute path created by the product). Registry lives under the operation's artifact dir (`.harness/operations/<id>/resources.json`), keyed to the OperationRecord: no PID-name or filesystem-glob inference.
2. **Classification.** Reconciliation classifies every candidate resource: `LIVE_OWNED` (operation non-terminal → preserve), `TERMINAL_ORPHAN` (operation durably terminal + registry ownership proof matches the exact identity → safe reconcile), `UNKNOWN_OR_UNOWNED` (no durable proof, another operation's owner, or identity mismatch → preserve + report). A campaign timeout, missing process or stale directory is not terminality.
3. **Safe reconciliation.** Workspace → `paseo workspace archive`; agent/session → inspect, stop if non-quiescent, then `paseo agent archive` (deterministic fixture sessions skipped); managed process → terminate exact registered handle/process group; staging root → remove only the exact registered path. Every action is idempotent: an already released/archived resource yields `alreadyReconciled`, never a new failure; real errors stay visible.
4. **Verification receipt.** Deterministic durable receipt (`resource-reconciliation.json`, the verification receipt) binds operation id, candidate digest, controller epoch, per-resource kind/identity/disposition/evidence, and reconciliation timestamps.
5. **Terminal + recovery.** SUCCEEDED/FAILED terminalization runs reconciliation after the terminal transition; a restart that loads an already-terminal operation with an incomplete registry runs the same deterministic sweep (crash between terminalization and cleanup cannot leave permanent uncontrolled residue). A bounded sweep at operation start reconciles proven terminal orphans in the same control root.
6. **Backpressure.** Generic policy ceilings (`operation resource policy`) for owned resources and concurrent provider sessions per operation enforced at registration/lease-time with a typed `RESOURCE_CEILING_EXHAUSTED` disposition; model output cannot widen a ceiling.

## Execution log (updated after every step)

- `2026-09-30` Reconstruction read complete: AGENTS.md; TARGET §15/§16; STATUS; CONFORMANCE; S14 WorkGraph (implementation + acceptance); S13 WorkGraph (resource amendment + AEH-V2-0130 investigation); LEDGER 0130/0131/0132/0133/0134/0135; `docs/evidence/s14/` summaries; `src/operations/controller.ts`; `src/operations/state.ts`; `src/runtime/providerLifecycle.ts`; `src/runtime/managed.ts`; `src/operations/portfolio.ts`; `tests/packed/s14ResourceStabilityCampaign.mjs`; `tests/packed/s13GovernedCampaignPolicy.mjs`. TARGET sha256 and HEAD verified. WorkGraph created before any implementation.
- `2026-09-30` C-W0 residue census (read-only, no kills): counts recorded under `docs/evidence/closure/` — see the W0 artifact row.

- `2026-09-30T19:0xZ` **C-W1/C-W2/C-W3 implemented (source frozen for campaigns after this log line):** new `src/runtime/operationResources.ts`; `src/operations/controller.ts` (workspace disposition + registration, terminal reconciliation, restart sweep/terminal-entry recovery); `src/operations/state.ts` (`workspaceDisposition` optional field only; the reconciliation receipt is the separate `resource-reconciliation.json` artifact); `src/runtime/providerLifecycle.ts` (provider-session ceiling); `src/candidates/direct.ts` (staging-root registration/release); `src/runtime/index.ts` export; `src/core/errors.ts` `RESOURCE_CEILING_EXHAUSTED`. `paseo workspace archive` was proven (disposable probe `wks_7602bfcdfcec0be1`) to remove the worktree, so terminal reconciliation snapshots bounded `.harness` evidence into the operation artifact dir before archiving; `tests/packed/s13GovernedOperationCampaign.mjs` verifies from the product snapshot when the worktree is gone (no check weakened). Closure harnesses: `tests/packed/closureResourceStabilityCampaign.mjs` (product reconciliation is the PASS mechanism; no campaign archive/stop/kill), `tests/packed/closureCampaignEnvelope.mjs`. Composite runtime artifacts now point at `docs/evidence/closure/`; `scripts/s14SelfHostingComposite.ts` and `scripts/s14AdversarialChecks.ts` updated. `npm run typecheck` exit 0; focused closure regressions 12/12; adjacent suites 103/103.
- `2026-09-30T19:1xZ` **C-W4 DONE** (12/12). Source freeze: no `src/**` change will be made while the real campaigns run. Candidate built: `release-1790795876290-870095-44d031fe`, buildDigest `834ab29bc89a79483f4d2d956736a802d7b47278d5d72e4a5a690c7070776c13` (gitSha `1d1dd97`, dirty). Canonical non-system suite: 205 files, 1,380 passed / 3 failed / 7 skipped — only the pre-existing `tests/paseoSdkResolve.test.ts` mise/Paseo environment drift (PRE_EXISTING_REPRODUCIBLE).
- `2026-09-30T19:2xZ` **C-W5 first real run `closure-resource-1` (2 iterations) truthful FAIL** (`docs/evidence/closure/resource-stability-product.json` sha256 `fa8d9c63e2bb89c55777da66401fcc928dd291c80fb22d1a4a1b7478e12ebcc6`): both operations terminal SUCCEEDED with product receipts `cleanupComplete: true` (11 dispositions each: 4 workspaces + 4 labeled/recorded agents + 2 staging roots already released inline), but the campaign census found three exact unproven residue classes: (1) the detached `operation monitor` process (not registered as a durable owned resource), (2) the pre-operation CHANGE route-triage `aeh-semantic-assessor-route` session (no operation label exists before the operation is created), and (3) the workspace record bound to that unlabeled assessor. Baseline did not return (1/1/1 then 3/2/2) → `OPEN_PRODUCT_DEFECT_EVIDENCED`. The failed run's staging root was removed by the harness (its Paseo records/processes now have no durable local proof and are classified UNKNOWN_OR_OWNED and preserved until proven).
- `2026-09-30T19:4xZ` **Product repair for the three classes:** `src/operations/monitorProcess.ts` registers the detached liveness monitor as a durable `managed-process` operation resource (`TERMINATE_ON_TERMINAL`); `src/semantic/runtime.ts` releases the exact single-turn assessor session immediately after the turn (and the exact assessor workspace when no operation context owns it — pre-operation triage); deterministic injected launches keep their fixture lifecycle. `npm run typecheck` exit 0; semantic/operations/lifecycle focused suites 100/100. Build `356fd7120bbe14b628660e1e068c469d508584f8e978f70bad7be2a75232ade3` (`release-1790797278234-956387-06149ed2`).
- `2026-09-30T19:5xZ` **C-W5 second real run `closure-resource-2` truthful FAIL** (`docs/evidence/closure/resource-stability-product.json` sha256 `93d2bcbc5d5a131a92d6d041ef51427b9421d2c39ead691e5aea58915d4674c4`): agents and processes returned to baseline 0/0 in both iterations (monitor + session fixes work; receipts `cleanupComplete: true`, each with the monitor `managed-process` reconciled). Remaining residue: 2 and 4 Paseo `local_checkout` workspace records created for the in-operation Semantic Assessor sessions (`candidate_impact`, `stack`) — the SDK label discovery path does not return `workspaceId`, so terminal reconciliation archived the assessor agents but not their per-agent local workspace records. Exact truth recorded; no campaign cleanup. 
- `2026-09-30T20:0xZ` **Product repair for the assessor workspace records:** `src/semantic/runtime.ts` now archives the exact per-agent `local_checkout` workspace returned by the assessor launch alongside the session (exact id from the launch result; `paseo workspace archive` is idempotent on archived ids; never the operation's worktree workspace). `npm run typecheck` exit 0; focused 38/38. Build `b73ed763afc36bf11ef9d69027ebdb6edaf40cc8830ee88c954f43bfc5ed1a55` (`release-1790798678838-1030000-1509ee85`).
- `2026-09-30T20:2xZ` **C-W5 third real run `closure-resource-3` truthful FAIL** (`docs/evidence/closure/resource-stability-product.json` sha256 `d0ac6e89029adb597ff3af5d97c617e3a4506b32a38464cb48d2c5f9b6075f38`): iteration 1 returned the baseline **0/0/0** with `cleanupComplete: true`; iteration 2 left exactly one worktree workspace (`wks_966e7dbb3cb38a9f`, the DIRECT reviewer attempt root `aeh-direct-…-T6a18e`). The reviewer agent `cbf925e4` WAS reconciled, but its workspace id was not present on the surfaces the product collected: the CLI launch returns only an explicit `--workspace` (undefined here), the SDK list entry omitted `workspaceId`, and the participant `executionBinding` was absent. Exact truth recorded; no campaign cleanup.
- `2026-09-30T20:3xZ` **Product repair for workspace-to-agent binding:** `src/runtime/operationResources.ts` uses the exact session inspection boundary (`inspectManagedPaseoAgent`) to resolve an operation-owned agent's Paseo workspace when the record/label surface does not carry it, then reconciles that workspace. Focused regression M added. `npm run typecheck` exit 0; focused 53/53. Build `43995128402098d38728cd5afdb4c138243476b461d8acd9fc19e306b2824177` (`release-1790800119836-1106184-f52427e3`).
- `2026-09-30T21:0xZ` **C-W5 fifth real run `closure-resource-5` truthful FAIL (retained as history):** iteration 1 terminal SUCCEEDED with product receipt `cd7d1c22…` and baseline 0/0/0; iteration 2 failed at supervisor initialization with a real provider/daemon flake (`AEH_OPERATION_SUPERVISOR_UNAVAILABLE` / `PASEO_PROVIDER_LEASE_TAKEOVER_BLOCKED` after a transient first attempt; Paseo daemon log shows client reconnect churn). The FAILED operation's receipt was not observed within the then-90s poll (the harness now polls 240s and the recovery-sweep path owns any missed receipt); baseline was still 0/0/0. Summary retained as `docs/evidence/closure/resource-stability-product.superseded-run5.json`. Run 6 follows on the same build (no source change).
- `2026-09-30T20:4xZ` **C-W5 fourth real run `closure-resource-4` superseded (truthfulness finding):** iteration 1 terminal SUCCEEDED with the complete 9-resource receipt (monitor + operation worktree + 3 agent-bound worktrees incl. the reviewer's + 3 agents + 2 inline staging roots) and baseline 0/0/0; iteration 2 terminal truthfully **FAILED** (`SEMANTIC_ASSESSMENT_UNAVAILABLE`, candidate-impact assessor provider `status=error`); its receipt still reconciled to 0/0/0. The lead's first harness gate accepted terminal-any-status and reported `PASS/CLOSED_BY_PRODUCT`; that gate reading was **too loose and is superseded**: the closure harness now requires `terminalRecord.status === SUCCEEDED` and lane `result === PASS` per iteration, and the superseded summary is retained under `docs/evidence/closure/resource-stability-product.superseded-run4.json`. In-operation assessor sessions were returned to their pre-existing behavior (terminal reconciliation owns them, now with inspect-resolved workspaces); only pre-operation triage sessions are released immediately, minimizing new daemon surface. Build `d266692df02c95bf432ac6c9578e127fb16584aa2cc18d469d2c844ed3b0a2ab` (`release-1790801178235-1161839-a11bfd3f`); run 5 follows.

- `2026-09-30T21:2xZ` **C-W5 PASS — `closure-resource-6` (strict gate).** `docs/evidence/closure/resource-stability-product.json` sha256 `94ed891de6e1d72b4039f7a733b55e6625a94a48df61fd5f0288147c5a157f14`: both real governed `change-direct` iterations terminal **SUCCEEDED** with lane `result: PASS`, product receipt `cleanupComplete: true` (receipts `d91dfadd…` / `fc47fbbb…`), product-reconciled workspaces/agents/processes to baseline **0/0/0** each iteration, `preservedIntact: true`, `postReconciliationNoGrowth: true`, `campaignSideCleanup: "NONE"`, classification `aehV2_0130/0135: CLOSED_BY_PRODUCT`. Superseded truthful failures: run-1 `fa8d9c63…`, run-2 `93d2bcbc…`, run-3 `d0ac6e89…` are recorded in this step log (their summaries were superseded in place); run-4 (`resource-stability-product.superseded-run4.json`) and run-5 (`…superseded-run5.json`) summary files are retained on disk. Build unchanged `d266692d…`.

## Evidence index

| Artifact | Path | sha256 |
| --- | --- | --- |
| Closure WorkGraph (this file) | `docs/CORE_ARCHITECTURE_V2_CLOSURE_WORKGRAPH.md` | (updated in place) |
| Residue census W0 | `docs/evidence/closure/residue-census.json` | `c352ae7eb9b99c86c4dd9f536a9345b3514accbbc321c2d466cf3b6129cadd49` |
| Focused regressions A–J (+K/L/M) | `docs/evidence/closure/focused-regressions.json` | `0d670ab4a9abd7bab7c2ed31f2c523427e703cb98b899874bb04843b6486087d` |
| Product stability campaign | `docs/evidence/closure/resource-stability-product.json` | `94ed891de6e1d72b4039f7a733b55e6625a94a48df61fd5f0288147c5a157f14` |
| Disposable campaign envelope | `docs/evidence/closure/disposable-self-modification-campaign.json` | `17e9b926cc17aaaafa34efa78222bcb1d6c1119f77b0e0375f398160d0f6e19a` |
| Lifecycle cancel/recovery rerun | `docs/evidence/closure/paseo-lifecycle-cancel-recovery.json` | `ead63493846fb3bb15b1ea7bba083a2486d89b642c62c187407d88ae22c73438` |
| Staleness analysis | `docs/evidence/closure/staleness-analysis.json` | `e6739566fcbc5880997dee543d42eab112587209e3e37de952a4c2d173024eec` |
| Adversarial/fail-closed checks | `docs/evidence/closure/adversarial-negative-checks.json` | `c100197d811fbf5e7cb7344f0157de22288cb809c19f916b42efae592a7a6a20` |
| Composite report | `docs/evidence/closure/self-hosting-composite-report.json` | `9b9e2134c6a4d6f22a0bbc2b6cd0b4c6a35202ced96da18b51c0f24a95d84b7f` |
| Composite manifest | `docs/evidence/closure/certification-evidence-manifest.json` | `ff1b5b113adeae34c88abdd6aaa5c2a4770f98a8d3b47cb8dd521def4ea54b9a` |
| Validation | `docs/evidence/closure/validation.json` | `572c2d1c1a944c3035206db2cb8337aa1ac92f8fc29f27afe9adc918a4b97da1` |

## Final record (closure lead; independent re-review pending)

- **Product result:** durable operation→resource ownership + terminal/recovery reconciliation is implemented in the product; terminal operations return owned resources to baseline without campaign-side cleanup, live/non-terminal resources are preserved, unknown/unowned resources are preserved and reported, reconciliation is idempotent and durable-receipted, crash/restart recovery reconciles proven terminal orphans, and generic policy ceilings bound owned resources/provider sessions with a typed exhaustion disposition.
- **Required architecture traceability:** registry/inventory (`src/runtime/operationResources.ts`, `.harness/operations/<id>/resources.json`), classification `LIVE_OWNED`/`TERMINAL_ORPHAN`/`UNKNOWN_OR_UNOWNED`, safe archive/stop/remove actions, durable receipt (`resource-reconciliation.json`), terminal integration (`terminalizeOperation` for SUCCEEDED/FAILED/CANCELLED), recovery (`executeOperation` terminal entry + `startDetachedOperation` sweep), backpressure (`RESOURCE_CEILING_EXHAUSTED`; `orchestration.operations.resources` policy).
- **Real re-verification:** stability `closure-resource-6` PASS (baseline 0/0/0 both iterations, `CLOSED_BY_PRODUCT`); disposable `closure-disposable-1` PASS (checkout byte-identical, 13/13 dispositions); lifecycle cancel/recovery 2/2 PASS — all on candidate buildDigest `d266692df02c95bf432ac6c9578e127fb16584aa2cc18d469d2c844ed3b0a2ab`.
- **Composite:** regenerated `ready: true` at policy digest `1be97df3…`; 21/21 required items, 47 verified artifacts, 0 failed; `readinessAuthority: DETERMINISTIC_COMPOSITE`, `modelAuthority: NONE`; product-reconciliation-backed resource item; byte-reproducible.
- **Remaining blockers/notes:** a fresh independent reviewer owns `SLICE_ACCEPTED`; the Campaign Director owns checkpointing. Post-v2 observations AEH-V2-0132/0133/0134, the `tests/paseoSdkResolve` mise drift and S14-F4 remain open and are out of this closure. Legacy pre-closure residue and unknown-ownership Paseo records were preserved and reported, never killed without durable proof. Two closure provider flakes are preserved as truthful failed-run history. No commit/push/stash/reset/clean was performed by this lead.

## Closure independent review corrections (2026-09-30)

The fresh independent closure reviewer returned `SLICE_ACCEPTED` (`docs/evidence/closure/review-closure.json` SHA-256 `e3055c207ccf1c463c08b4e1aa13dbb1a6de254d2ed88eedca6430b55af4d110`) and required pre-checkpoint documentation corrections, applied here without any product-source or re-verification change:

- **F1 (documentation overclaim):** the terminal-record `resourceReconciliation` summary-field claims are corrected — the verification receipt is the durable `resource-reconciliation.json` artifact (with `cleanupComplete` and per-resource dispositions); `state.ts` carries only the `workspaceDisposition` optional field.
- **F2 (retention precision):** the "five superseded runs retained" wording is corrected to the retained run-4/run-5 summary files plus the run-1..3 digest/history entries in the step log.
- **F3 (receipt attribution):** the STATUS cancel/recovery sentence no longer attributes product reconciliation receipts to the lifecycle lane; its asserted evidence is cancel/recovery state.
- **F6 (row-geometry citation):** the staleness bounded-scope reasoning for `multi-worker` cites the row's own geometry (2 work units / 1 wave; the generic provider-session ceiling is 16 and cannot be reached) and `distributed-execution` cites its lease counts (prepared=2 / released=2, far below 16), not observational anecdote.
- **F4/F5 (non-blocking, post-v2):** harden `assertProviderSessionCapacity` fail-open/check-then-acquire behavior and disclose the campaign-scope census boundary; both recorded as post-v2 hardening candidates and neither affects this acceptance.
