# Core Architecture v2 Status

**Role:** Living roadmap and current-state evidence map. The normative target is [Core Architecture v2](CORE_ARCHITECTURE_V2.md). Conformance lane boundaries are in [Core Architecture v2 Conformance](CORE_ARCHITECTURE_V2_CONFORMANCE.md); historical incidents and findings remain in the [Engineering Ledger](ENGINEERING_LEDGER.md).

## Current campaign disposition — 2026-10-01

**Core Architecture v2 is complete.** S1–S12 are accepted, including the S10 security/isolation/SAST gate. The S10 acceptance gap is closed by the fresh independent `SLICE_ACCEPTED` review at [review-s10-final.json](evidence/s10/review-s10-final.json), reviewed against commit `87144f77a0c99088d385e848925643b5439422ea`. S13 is independently accepted with its complete 22-row capability matrix (21 current rows and the Control Center browser-only row marked `NOT_APPLICABLE`); S14 and final resource closure are independently accepted; CI Alignment is accepted with its required hosted matrix green.

This closes the Core-v2 architecture acceptance gates. It does not close unrelated post-v2 observations: AEH-V2-0132, AEH-V2-0133, AEH-V2-0134, the `tests/paseoSdkResolve` environment drift, and S14-F4 fixture hygiene remain separately recorded and nonblocking. OPA live evaluation and rootless Podman/OCI provider availability are not claimed by S10.

## Evidence and verification boundary

The baseline current-state evidence came from the read-only architecture audit at commit f1518caf405f44851227cf73ea37b79267b1a098. S1–S8 have been accepted on their deterministic slice gates; the Campaign Director accepted S8 on 2026-09-25. S7/S8 deterministic results do not certify live providers, browsers, OPA, adversarial behavior, or packed end-to-end journeys. On 2026-09-25, S10 exercised a real rootless bwrap isolation provider and a real local OSS Trivy v0.70.0 scanner; those are local real-provider executions and do not certify hosted providers, rootless Podman/OCI, OPA, packed journeys, or S13 provider certification. On 2026-09-26, S11 produced real OpenAPI/BDD contract evidence, a real loopback service integration lifecycle, and a real Playwright Chromium Control Center browser/visual journey through the candidate-bound validator path; those are local BROWSER/VISUAL and SYSTEM_DETERMINISTIC executions with the S9 file-scripted Paseo provider boundary and do not certify hosted providers, live models, packed UI journeys, or S13 provider certification. The first S11 submission was independently reviewed and rejected; the repaired slice requires specialized lane evidence to be produced only by the matching provider execution, binds VISUAL evidence to its committed reference baseline and comparison configuration, makes each lane's browser artifacts durable and campaign-verified, and computes a candidate digest the campaign cannot invalidate. S11 was independently re-reviewed by a fresh reviewer (R3) and accepted on 2026-09-26 across all 18 gate items; the S11 checkpoint is commit 78de9c5. On 2026-09-26, S12 produced candidate/execution-bound telemetry identity, real local metric instruments, a versioned advisory eval corpus, and a local OTLP loopback export verification; the OTLP lane is a local in-process receiver, not a hosted collector, and no REAL_PROVIDER claim is made.

EVIDENCE LANE names the kind of evidence cited in the audit, not a claim that a test or provider was rerun during S0. REAL_PROVIDER, BROWSER, ADVERSARIAL, and PACKED_E2E evidence remain separate from CONTRACT, UNIT, and INTEGRATION evidence. Historic test counts are checkpoint records, not current certification.

## PR #64 CI alignment evidence — 2026-10-01

The dedicated Core Architecture v2 CI alignment campaign is recorded in [CORE_ARCHITECTURE_V2_CI_ALIGNMENT.md](CORE_ARCHITECTURE_V2_CI_ALIGNMENT.md). Baseline PR run 36782724704 was reconstructed at head bbe2bc6624d2b1a51f001f6dec566786590b1296. Historical checkpoint 325605ebf842d098aa0fdbe556c433c3aac9cc0b was tested by run 36803071091: 11 jobs passed; packaged-consumer failed at npm pack output capture and rootless-isolation failed on Ubuntu 24.04 namespace/validator execution. At the pre-amendment snapshot on 2026-10-01, PR #64 head was 48276ea842fa3af475eb32632d995b18f755f17c; run 36817138377 passed 12/13 jobs. Rootless-isolation passed the real hosted campaign 3/3 on Ubuntu 22.04/bwrap 0.6.1 with the non-root namespace preflight; the remaining packaged-consumer failure was P3, where the clean consumer lacked paseo and @getpaseo/client for the configured session-scoped Semantic Assessor. The production refusal to drop session identity was correct; the missing requirement was a typed deterministic INTENT response in the test fixture. The typed P3 fixture repair passed a direct built-CLI smoke locally. As of this snapshot, it was the next checkpoint candidate and had no hosted result. The Ubuntu 22.04 runner pin is temporary during GitHub's deprecation window; migrate before its scheduled 2027-04-17 full unsupported date ([runner-image schedule](https://github.com/actions/runner-images/issues/14254)). No Core-v2 runtime gate was weakened. Fresh independent review and green checks for the amended SHA remain required.

This is CI-alignment evidence, not new Core Architecture certification. Historical pre-amendment snapshot (2026-10-01): PR #64 head was `48276ea842fa3af475eb32632d995b18f755f17c`; run 36817138377 passed 12 jobs and failed only packaged-consumer at P3. The historical checkpoint 325605/run 368030 failures remain recorded above. The typed P3 fixture repair was the next checkpoint candidate and had no hosted result as of that snapshot. Fresh independent review and green hosted checks for the amended SHA remain required. The normative TARGET remains unchanged.

Intermediate hosted snapshot: PR #64 head `ff2b9f279fcf6c974f626875dc99248881613a79`; GitHub run 36822314115 checked PR merge SHA `64e3f963f7bf70e287da13f0f9aefb13e26c0450` for this head and completed 12/13 jobs. Packaged-consumer, rootless-isolation, Pact, and the other specialized lanes passed; test/release failed only at F5, where `tests/security/isolation.test.ts` timed out after 5,000ms while awaiting host capability detection. The generic Ubuntu 24.04 runner image includes Podman, and production capability detection can invoke its `info` command; the hosted log confirms the timeout at the detection call but does not identify the exact subprocess. The test now uses a unique temporary root and empty executable PATH, marks the suite sequential, and passes locally 10/10 in 417ms. This local repair was later verified by run 36824682741 on implementation checkpoint d21d6b5, before the current documentation-only amendment. See the CI inventory for the full historical and current job matrices.

Last fully verified campaign checkpoint before this status wording reconciliation (2026-10-01): PR #64 was OPEN at head `963cc3010dc5b372a282da4f37859a8ce6a27796`. GitHub Actions run [36832161124](https://github.com/JamesMorales04/agentic-engineering-harness/actions/runs/36832161124) completed all 13 required jobs successfully, including test/release, human-e2e, adversarial-system, rootless-isolation, packaged-consumer, full-stack-contract, supply-chain, provider contracts, Pact, and OCI integration. Round 16 independently accepted this status-doc candidate. The exact-SHA result for this final wording snapshot is attached to its PR Actions run. No merge-readiness claim is made here.

**Current architecture disposition:** S10 final re-acceptance is `SLICE_ACCEPTED` and the S10 acceptance gap is closed. S1–S12, the S13 22-row matrix, S14, final resource closure, and CI Alignment are accepted; Core Architecture v2 is complete. The separate follow-up release/docs reconciliation still awaits its own independent review and ordinary PR controls. AEH-V2-0132/0133/0134, `tests/paseoSdkResolve` environment drift, and S14-F4 remain nonblocking post-v2 observations.

## Stable subsystem roadmap

| Subsystem | TARGET | CURRENT | STATUS | GAP | DEPENDENCIES | NEXT GATE | EVIDENCE | EVIDENCE LANE | LAST VERIFIED |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Candidate truth and assembly | Immutable CandidateRevisions; controller-owned assembly; truthful bases and lineage. | Assembly, wave lineage, repair, and workspace identity checks are wired. S1 candidate assembly invalidates bindings and structured results for the prior candidate. S4 impact and assurance use the exact assembled CandidateRevision; S6 acceptance and objective-completion evidence bind that current candidate. S7 provenance binds the same candidate. S8 controller bootstrap provisions an isolated internal execution workspace before acceptance while binding it to the current candidate; the delivery branch is separate from the local candidate branch. | S7 deterministic gate accepted 2026-09-24; S8 deterministic gate accepted by Campaign Director 2026-09-25. | No known deterministic S8 candidate-identity gap. Real-provider workspace and packed-candidate behavior remain separate evidence lanes. | S1 current CandidateRevision and BuildIdentity; S6 AcceptanceOracle before effects; frozen supply-chain policy. | Preserve current CandidateRevision and BuildIdentity; keep provider and packed certification pending. | src/candidates/assembler.ts, src/core/run.ts, src/architecture/candidateAssurance.ts, src/provenance/generate.ts, src/operations/controller.ts, src/delivery/finalize.ts; tests/candidateIdentity.test.ts, tests/deliveryFinalize.test.ts, tests/security/toolActionGate.test.ts. | CONTRACT; UNIT; INTEGRATION; BUILD; SYSTEM_DETERMINISTIC; REAL_PROVIDER/BROWSER/OPA_PROVIDER/ADVERSARIAL/PACKED_E2E pending | S8 remediation evidence is recorded below. |
| Authority kernel | Frozen policy, role ceilings, leases, epoch fencing, gated actions, and consumed scoped HumanDecisions. | Controller mutations retain current epoch, controller token, candidate and frozen-policy checks. ToolActionGate classifies paseo.workspace.create as LOCAL_RESOURCE_CREATION and excludes it from delivery external-effect policy; the fenced controller still persists intent/receipt and unresolved workspace identity remains HUMAN_REQUIRED. Public GitHub, push and PR actions remain current-acceptance gated. | S3 and S8 deterministic authority/action gates accepted; S8 by Campaign Director 2026-09-25. | No known deterministic authority/action gap from S8. Live provider identity inspection and external certification remain pending. | Candidate identity; S1 execution identity; current policy and epoch. | Preserve S3 authority/fencing and accepted S8 controls; retain separate provider, browser, OPA and packed evidence lanes. | src/security/actionPolicy.ts, src/security/toolActionGate.ts, src/security/gatedAction.ts, src/security/actionReconciliation.ts, src/operations/controller.ts; tests/security/actionPolicy.test.ts, tests/security/toolActionGate.test.ts, tests/security/gatedAction.test.ts, tests/security/actionReconciliation.test.ts. | CONTRACT; UNIT; INTEGRATION; BUILD; SYSTEM_DETERMINISTIC; REAL_PROVIDER/BROWSER/OPA_PROVIDER/ADVERSARIAL/PACKED_E2E pending | S8 remediation evidence is recorded below. |
| Intent, route, assurance, and ResolvedOperationPolicy | Semantic intent and route constrained by deterministic policy; independent assurance minimum; frozen policy digest. | S1 persists the immutable ResolvedOperationPolicy before participant binding. Non-explicit intent and route use typed Semantic Assessor judgments through the AgentTopology-resolved Paseo runtime; explicit user intent remains DETERMINISTIC. INTENT admits HIGH-risk requests within existing assessment ceilings; deterministic high-risk routing and CRITICAL assurance floors remain unchanged. S3 consumes frozen policy identity at sensitive-action, product-choice, cancellation, and decision boundaries without changing S2 intent or route semantics. | IMPLEMENTED | REAL_PROVIDER execution is not established by deterministic fakes. | S1 identity; AgentTopology and semantic capability policy; Paseo runtime; S3 authority kernel. | Preserve S1/S2 policy and deterministic floors; certify provider behavior separately. | src/entry.ts, src/audit/intent.ts, src/core/triage.ts, src/cli.ts, src/operations/change.ts, src/core/contract.ts, src/semantic/assessment.ts, src/security/toolActionGate.ts; tests/auditIntent.test.ts, tests/triage.test.ts, tests/semanticAssessment.test.ts, tests/security/toolActionGate.test.ts. | CONTRACT; UNIT; REAL_PROVIDER pending | 2026-09-24 S3 policy-consumption regression verification; S2 semantics unchanged |
| Semantic decision runtime | Production semantic decisions use bounded evidence-bound model assessment with deterministic post-checks. | SemanticAssessmentServiceV1 resolves one AEH Semantic Assessor from AgentTopology and semantic capability policy, executes the configured agent through Paseo, validates typed output, policy, binding, receipt, cache, and actual session provenance, and stores SemanticAssessmentV1. INTENT, ROUTE, STACK, ISSUE, FAILURE, CANDIDATE_IMPACT, and VALIDATION_NEED are supported. Semantic Assessor is an AEH Agent and is rejected from WorkGraph Participant invocation. STACK additionally requires canonical-root and repository-content binding; its generic inventory is aligned with the Git digest file set and currentness is checked before/after capture and after assessment. | IMPLEMENTED | Deterministic Paseo fakes verify contracts but do not establish REAL_PROVIDER behavior. | Accepted S1 identity interfaces; AgentTopology; Paseo; deterministic semantic capability policy. | Keep provider certification separate; reject stale topology, session, evidence, policy, and cache provenance. | src/semantic/assessment.ts, src/semantic/runtime.ts, src/agents/permissions.ts, src/agents/outputContracts.ts, src/workers/agentPrompt.ts; tests/semanticAssessment.test.ts, tests/semanticAssessmentRuntime.test.ts, tests/semanticAssessorTopology.test.ts, tests/participantsStackDiscovery.test.ts. | CONTRACT; UNIT; REAL_PROVIDER pending | 2026-09-24 S2 boundary remediation deterministic verification |
| Stack discovery and semantic stack profile | HYBRID: model interprets bounded repository/candidate evidence; deterministic code verifies identity, read boundary, receipts, profile schema, provenance, policy, and cache freshness. | runTask supplies current bound repository/candidate evidence to discoverProjectStackProfile; profile output is MODEL-only, preserves unknowns, and is passed to delegated participant planning and validation script resolution. STACK requires `repositoryRootDigest` for the canonical realpath and matches `repositoryDigest` before scan, after evidence capture, after the assessor returns, and before profile return. Evidence paths come from the Git-tracked and non-ignored untracked inventory used by the digest, or a stricter generic subset; ignored `.env` and `.harness` runtime/cache paths are absent. Model-proposed projectSkillRoots are preserved only after normalized relative-path, existence, and canonical-root containment checks and grant no skills/capabilities. Technology-specific deterministic recognition and detector APIs are removed. | IMPLEMENTED | Repository evidence limits may produce explicit unknowns. Real model interpretation is not certified by fixture tests. | Semantic decision runtime; candidate identity; authorized repository boundary. | Keep the generic non-ignored read inventory, root/content checks, and stale-result rejection; actual provider evidence remains a separate lane. | src/participants/stack.ts, src/participants/index.ts, src/core/run.ts, src/core/git.ts, src/semantic/runtime.ts, src/agents/waveExecutor.ts, src/architecture/validationRequirements.ts; tests/participantsStackDiscovery.test.ts, tests/participantsCanonical.test.ts, tests/semanticAssessmentRuntime.test.ts. | CONTRACT; UNIT; REAL_PROVIDER pending | 2026-09-24 S2 repository evidence/path remediation deterministic verification |
| Issue intake | HYBRID: typed ISSUE interpretation plus canonical Planner authoring and deterministic source/scope/policy checks. | The public issue implement/run path prepares the existing issue, then starts a managed controller run. Controller-owned disposable workspace setup remains pre-acceptance; default auto-handoff no longer creates a public branch before acceptance. Accepted finalization gates and receipts any missing issue/branch and configured push/PR effects. The standalone `aeh sdd handoff` command reports `SDD_HANDOFF_UNSUPPORTED` before project loading instead of entering delivery without managed context. | S2 deterministic gate accepted; S8 issue-flow and CLI-surface remediation accepted by Campaign Director 2026-09-25. | No known deterministic issue-flow gap in the managed path. Standalone SDD handoff is intentionally unsupported; live GitHub/Paseo execution is not certified by fixtures. | Semantic decision runtime; immutable issue snapshot; canonical Planner; route policy. | Preserve the managed issue path and acceptance boundary; keep live provider and packed issue journeys pending. | src/issues/intake.ts, src/issues/workflow.ts, src/cli.ts, src/delivery/handoff.ts, src/delivery/finalize.ts; tests/issueIntake.test.ts, tests/issueDeliveryIdentity.test.ts, tests/deliveryFinalize.test.ts, tests/cliS8Handoff.test.ts. | CONTRACT; UNIT; INTEGRATION; BUILD; REAL_PROVIDER/PACKED_E2E pending | S8 remediation evidence is recorded below. |
| WorkGraph, scheduling, and resource claims | Model proposes decomposition; deterministic validation/scheduler enforces dependencies, scope, claims, and isolation. | WorkGraph, claims, topological waves, parallelism, and scope checks remain deterministic. Typed semantic ROUTE and issue-route decisions feed contract routing and the bounded Planner/WorkGraph path; stack profile is supplied to ParticipantPlan compilation and validation resolution. S4 selects only a final-wave CandidateImpact whose candidate identity exactly equals the operation's current CandidateRevision; missing or stale impact blocks the assurance gate. S6 terminal success also checks that every required WorkUnit is accounted for in the complete current objective evidence set. | IMPLEMENTED | Live semantic-provider behavior and packed operation journeys remain separate pending lanes; S4 alone does not make validation or review an AcceptanceOracle. S6 deterministic implementation was accepted by the Campaign Director on 2026-09-24; provider and packed evidence remain separate pending lanes. | Route/policy; candidate truth; semantic decision runtime; S4 assurance compiler. | Preserve deterministic claims and current candidate bases; require accounted WorkUnits and the current AcceptanceOracle/objective-completion checks for terminal success. | src/architecture/workGraph.ts, src/agents/parallelism.ts, src/agents/waveExecutor.ts, src/core/run.ts, src/issues/intake.ts; tests/architectureWorkGraph.test.ts, tests/parallelism.test.ts, tests/triage.test.ts, tests/candidateWaveLineage.test.ts. | CONTRACT; UNIT; INTEGRATION; REAL_PROVIDER pending | 2026-09-24 S4 focused 7 files / 41 tests; full non-system 153 files passed, 1 skipped (848 passed, 7 skipped); typecheck/build passed. |
| Participant and role compiler | Deterministically compile minimal ParticipantPlan, canonical roles, specializations, skills, ToolPacks, scope, and output contracts. | S1 role ceilings and per-participant bindings remain unchanged. Semantic Assessor is included only in the Agent role space and cannot be selected as a Participant; stack profile specializes bounded planner assignments without granting new tools or authority. | PARTIAL | Repairer selection and complete catalog/tool-policy consumption remain open. | WorkGraph; policy; KnowledgeGate; Semantic Assessor remains outside Participant plans. | Retain deterministic S1 role ceilings and Agent/Participant distinction. | src/architecture/participantPlan.ts, src/participants/types.ts, src/agents/types.ts, src/agents/config.ts, src/workers/agentPrompt.ts; tests/semanticAssessorTopology.test.ts, tests/architectureWorkGraph.test.ts. | CONTRACT; UNIT | 2026-09-24 S2 deterministic verification |
| Knowledge, Librarian, and skill trust | Version-aware knowledge resolution; read-only Librarian; deterministic trust; operation-local skill procedures delivered and identity-bound. | S1 retains accepted ephemeral procedure text and per-step claim/source grounding in a scoped SkillManifest, projects it verbatim only to the authorized assignee, and binds its digest through launch and result provenance. The procedure-text mutation rejection and trust-decision-digest mutation rejection both remain asserted. | PARTIAL | Provider-backed Librarian evidence and full freshness/source-policy lifecycle remain unverified; no real-provider claim is made. | Participant compiler; S1 SkillManifest identity; S5 context runtime. | Keep trust deterministic and preserve the exact S1 manifest/projection boundary; provider evidence remains a separate lane. | src/knowledge/index.ts, src/architecture/executionIdentity.ts, src/workers/agentPrompt.ts; tests/knowledgeGate.test.ts, tests/agentPromptStructuredResultProvenance.test.ts, tests/distributedExecutionIdentity.test.ts. | CONTRACT; UNIT; REAL_PROVIDER pending | 2026-09-24 S1 remediation deterministic verification |
| Context Runtime and progressive exposure | JIT, authorized, budgeted, source-provenanced ContextManifest; verbatim normative fragments; exact prompt digest. | S5 wires strict controller-only ContextRefAuthorizationV1 issuance from current S3 owner state and a current participant read lease. Grants bind operation/revision, project, candidate, participant/generation, blueprint, policy, epoch, actual session, exact refs/source digests, expiry, and retrieval budget. Retrieval rechecks the current binding, envelope, source bytes, session, expiry, request uniqueness, and durable limits; continuations bind the current binding, references, and receipts and reject stale/replayed state. S1 launch ContextManifest/PromptManifest digests remain immutable and authoritative; StructuredResult adds a ProgressiveContextManifestV1 digest over the exact receipt set, delivered-byte digests, used budget, state, and launch digests. Planner/Explorer handoffs use their existing deterministic projections; normative and accepted skill content remain verbatim. At grant issuance, the controller verifies the immutable launch ContextManifest digest against the current ExecutionBinding, its envelopeDigest against the verified durable envelope provenance digest, and exact unique (refId, artifactPath, sourceDigest) set equality before constructing the grant. | DETERMINISTICALLY_VERIFIED (S5 gate accepted) | No known S5 deterministic gate gap. REAL_PROVIDER behavior for live Paseo/MCP context retrieval remains unverified. | Accepted S1 identity; accepted S3 owner/epoch and lease semantics. | Preserve the frozen grant and receipt checks; provider-backed context and session behavior remain separate evidence lanes. | src/context/authorizationV2.ts, src/context/runtimeV2.ts, src/context/retrieval/, src/context/envelope.ts, src/context/gateway.ts, src/workers/agentPrompt.ts; tests/contextAuthorizationV2.test.ts, tests/contextRuntimeV2.test.ts, tests/contextEfficiency.test.ts, tests/agentPromptStructuredResultProvenance.test.ts. | CONTRACT; UNIT; INTEGRATION; BUILD; SYSTEM_DETERMINISTIC; REAL_PROVIDER pending | 2026-09-24 S5 remediation red/green 1 file (4 failed / 4 passed before repair; 8 passed after); focused context/session/provenance 12 files / 125 tests; full non-system 152 files passed, 1 skipped (835 passed, 7 skipped); independent Campaign Director rerun: focused 12 files / 125 tests, TypeScript exit 0, build exit 0 (`release-1790294205611-17-b4fe25cb`), relevant deterministic system suites 3 files / 45 tests; REAL_PROVIDER pending. |
| ExecutionBlueprint and RoleInvocation identity | Complete static blueprint and per-launch binding without granting authority. | Versioned ExecutionBlueprint V2 and S1 ExecutionBinding V2 remain authoritative. S5 issues addressable context only after checking the exact current participant binding, candidate, policy, controller epoch, actual session, and an existing read lease; context cannot add or widen lease/capability authority. The prepared launch prompt and context digests remain those bound in ExecutionBinding. Before grant construction, the issuer also checks the exact bound launch-manifest digest, envelope provenance digest, and full addressable-ref tuple set. | DETERMINISTICALLY_VERIFIED (S1 contract plus S5 context/session lifecycle) | No known S5 deterministic identity gap; REAL_PROVIDER execution remains unverified. | Policy; ParticipantPlan; SkillManifest; accepted S3 owner and lease contracts. | Preserve lifecycle-owned binding transitions and keep provider behavior in a separate evidence lane. | src/architecture/executionIdentity.ts, src/security/executionLease.ts, src/context/authorizationV2.ts, src/workers/agentPrompt.ts; tests/contextAuthorizationV2.test.ts, tests/agentPromptStructuredResultProvenance.test.ts, tests/security/controllerFencing.test.ts. | CONTRACT; UNIT; INTEGRATION; REAL_PROVIDER pending | 2026-09-24 S5 remediation verification: context/session/provenance 12 files / 125 tests; full non-system 152 files passed, 1 skipped (835 passed, 7 skipped); TypeScript/build passed; relevant deterministic system suites 3 files / 45 tests passed. |
| StructuredResult provenance | Every result binds current candidate, operation execution revision, generation, epoch, blueprint, policy, context/prompt, skills, runtime, and contract. | S1 full binding checks remain in force. S5 closes retrieval at result acceptance and persists ContextRetrievalEvidenceV1 with exact current binding and launch-manifest digests, durable receipt IDs/digests, delivered-byte digests, budget/use, and a domain-separated ProgressiveContextManifestV1 digest. Result reload verifies this evidence against durable receipts and rejects stale or altered evidence. The controller only issues the launch grant after exact ContextManifest/envelope evidence equality has been established. | DETERMINISTICALLY_VERIFIED (S1 adapter plus S5 context evidence) | Deterministic adapters do not certify live provider/MCP behavior; no known S5 deterministic evidence gap. | S1 execution identity; S3 controller fencing; S5 current context authorization. | Preserve full provenance and receipt closure; real transport/provider behavior remains a separate evidence gate. | src/workers/resultGateway.ts, src/context/authorizationV2.ts, src/workers/agentPrompt.ts; tests/structuredResultGateway.test.ts, tests/agentPromptStructuredResultProvenance.test.ts, tests/contextAuthorizationV2.test.ts, tests/distributedExecutionIdentity.test.ts. | CONTRACT; UNIT; INTEGRATION; REAL_PROVIDER pending | 2026-09-24 S5 remediation verification: context/session/provenance 12 files / 125 tests; full non-system 152 files passed, 1 skipped (835 passed, 7 skipped); TypeScript/build passed; relevant deterministic system suites 3 files / 45 tests passed. |
| Paseo session identity | Reuse requires exact durable operation/participant/candidate/blueprint/policy/context binding. | PaseoSessionBindingV1 now has strict full canonical identity including operation execution revision, participant generation, candidate, blueprint, policy, context/prompt digests, and controller epoch. Worker prompt preparation durably binds the actual provider session; only ACTIVE exact matches are reusable. A different actual session rotates the durable binding; same-session semantic drift rejects reuse until a new session is materialized. StructuredResult still names the actual returned session. | DETERMINISTICALLY_VERIFIED (S5 binding and lifecycle) | No known S5 deterministic gate gap. REAL_PROVIDER service behavior is not certified by fakes. | S1 ExecutionBinding; S5 context manifests; accepted S3 epoch fencing. | Keep exact equality at every reuse boundary; provider-backed session behavior remains separate evidence. | src/paseo/sessionBinding.ts, src/workers/agentPrompt.ts, src/paseo/launchSpec.ts, src/workers/resultGateway.ts; tests/paseo/sessionBinding.test.ts, tests/agentPromptStructuredResultProvenance.test.ts, tests/paseoPendingExecutionIdentity.test.ts, tests/paseoInitialTurnBarrier.test.ts. | CONTRACT; UNIT; INTEGRATION; REAL_PROVIDER pending | 2026-09-24 S5 remediation verification: session/context/provenance 12 files / 125 tests; full non-system 152 files passed, 1 skipped (835 passed, 7 skipped); TypeScript/build passed; relevant deterministic system suites 3 files / 45 tests passed. |
| CandidateImpact | Assess the assembled candidate semantically and deterministically with typed impact dimensions. | Every production CandidateAssembler path in runTask, wave integration, and repair assesses bounded post-assembly file/patch receipts and emits candidate, base-candidate, patch, changed-path, semantic digest, and impact digest bindings. runTask consumes the exact current impact and recompiles the pure deterministic assurance plan under the current frozen policy. S6 consumes the resulting assertions/evidence-strength without reinterpreting impact. Invalid, absent, stale, or tampered impact fails closed. | DETERMINISTICALLY_VERIFIED (S4 deterministic gate; provider pending) | MODEL impact interpretation and live provider/session behavior are not certified by deterministic fixtures. S6 does not widen or rewrite the S4 impact/assertion/evidence-strength interface. | Candidate truth; accepted S2 semantic runtime; S3 policy/authority; S5 context/session identity. | Preserve post-assembly binding and fail-closed behavior; acceptance consumes only assertions compiled for the exact current impact and candidate. | src/candidates/assembler.ts, src/candidates/wave.ts, src/candidates/repair.ts, src/core/run.ts, src/architecture/candidateAssurance.ts, src/architecture/acceptanceOracle.ts; tests/candidateAssembler.test.ts, tests/candidateAssurance.test.ts, tests/candidateWaveLineage.test.ts, tests/runRepairCandidateLifecycle.test.ts, tests/candidateRepairProtection.test.ts, tests/acceptanceOracle.test.ts. | S4 CONTRACT; S4/S6 UNIT; INTEGRATION; REAL_PROVIDER pending | 2026-09-24 S4 baseline retained; S6 consumption focused suite 11 files / 124 tests, full non-system 157 files passed / 1 skipped (931 passed / 7 skipped), typecheck/build passed; no REAL_PROVIDER claim. |
| Review and assurance recompilation | Impact and policy compile required reviewers, dimensions, and independent evidence. | S4 compiles exact configured read-only Reviewer identities for required dimensions and invokes those assignments even when ordinary review is disabled or the route is DIRECT. Each assignment emits candidate/impact/policy-bound evidence from the exact Reviewer session identity; failed or malformed output fails the report. S6 additionally requires the actual current participant ExecutionBinding/session and persisted result artifact before including Reviewer evidence. Agreement remains evidence only, not acceptance. Repair/rollback recompiles for the resulting current candidate. | DETERMINISTICALLY_VERIFIED (S4 deterministic gate; provider pending) | Live reviewer/provider behavior is not certified. S6 deterministic fixtures cover current session/provenance consumption; they do not certify live Reviewer behavior. | CandidateImpact; ParticipantPlan; frozen policy; S6 acceptance. | Keep assignments backed by actual independent current sessions; consume their evidence only through the AcceptanceOracle. | src/agents/reviewLifecycle.ts, src/core/run.ts, src/architecture/candidateAssurance.ts, src/architecture/acceptanceOracle.ts; tests/candidateAssurance.test.ts, tests/reviewLifecycle.test.ts, tests/reviewRepairCandidateLifecycle.test.ts, tests/runRepairCandidateLifecycle.test.ts, tests/acceptanceOracle.test.ts. | CONTRACT; UNIT; INTEGRATION; REAL_PROVIDER pending | 2026-09-24 S4 baseline retained; S6 consumption focused suite 11 files / 124 tests, full non-system 157 files passed / 1 skipped (931 passed / 7 skipped), typecheck/build passed; no REAL_PROVIDER claim. |
| Validation | Resolve required properties to approved validators; produce machine-verifiable current-candidate facts. | S4 deterministically maps impact dimensions to typed validation requirements, resolves them under frozen policy, and executes the approved project/configured action or fails closed when it is absent, disallowed, unsupported, or non-PASS. Checks bind candidate, impact digest, policy digest, requirement, and underlying check. S6 consumes resolved checks as evidence for their mapped assertions; validation PASS remains necessary evidence where required, never product acceptance by itself. | DETERMINISTICALLY_VERIFIED (S4 deterministic gate; provider pending) | Live provider, browser, and broader certification coverage remain pending; local validation and S4 PASS are not product acceptance. | WorkGraph; CandidateImpact; provider contracts; semantic stack profile; frozen policy. | Keep S4 requirements fail-closed and evidence candidate-bound; S6 requires the complete resolved evidence set and an accepted independent oracle disposition. | src/core/verify.ts, src/core/run.ts, src/architecture/validationRequirements.ts, src/architecture/candidateAssurance.ts, src/architecture/acceptanceOracle.ts, src/validators/external.ts; tests/validationProviders.test.ts, tests/candidateAssurance.test.ts, tests/candidateAssembler.test.ts, tests/runRepairCandidateLifecycle.test.ts, tests/acceptanceOracle.test.ts. | CONTRACT; UNIT; INTEGRATION; REAL_PROVIDER and BROWSER pending where required | 2026-09-24 S4 baseline retained; S6 consumption focused suite 11 files / 124 tests, full non-system 157 files passed / 1 skipped (931 passed / 7 skipped), typecheck/build passed; no live-provider claim. |
| AcceptanceOracle | Assertions, verification requirements, deterministic resolver, provenance evidence, independent acceptance oracle, required certification. | S6 still persists the current controller-evaluated ACCEPTED/REJECTED disposition bound to operation/candidate/policy/execution/controller epoch. S8 starts GitHub issue/branch/push/PR delivery only after current acceptance; objective completion recognizes accepted HANDOFF_ONLY when handoff is requested without push/PR finalization. Standalone `aeh sdd handoff` is explicitly unsupported and cannot trigger acceptance or delivery. | S6 deterministic gate accepted; S8 deterministic integration and CLI-surface remediation accepted by Campaign Director 2026-09-25. | No known deterministic acceptance-boundary gap. Live Lead/Reviewer/provider journeys and complete packed evidence remain pending. | Identity; impact; validation; review; human decisions. | Preserve current-candidate oracle and objective-completion gates without broadening S6 semantics. | src/architecture/acceptanceOracle.ts, src/architecture/objectiveCompletion.ts, src/core/run.ts, src/delivery/finalize.ts, src/delivery/handoff.ts; tests/acceptanceOracle.test.ts, tests/objectiveCompletionLifecycle.test.ts, tests/deliveryFinalize.test.ts, tests/cliS8Handoff.test.ts. | CONTRACT; UNIT; INTEGRATION; BUILD; SYSTEM_DETERMINISTIC; REAL_PROVIDER/BROWSER/OPA_PROVIDER/ADVERSARIAL/PACKED_E2E pending | S8 remediation evidence is recorded below. |
| Repair, recovery, and convergence | Typed diagnosis and bounded repair/replan; every repair creates a new candidate. | Production runTask preserves mechanically certain failure facts as DETERMINISTIC and sends ambiguous bounded failure evidence to the candidate-bound Semantic Assessor. Repair assembly carries post-assembly CandidateImpact into S4 assurance recompilation; quality repair/rollback revalidate and recompile for the current candidate. S6 consumes only assertions/evidence strength for that current candidate; prior-candidate acceptance evidence cannot support a repaired candidate. | PARTIAL | Broader recovery/quality convergence and live semantic-provider behavior remain unverified; S6 acceptance ordering and current-candidate checks are covered by deterministic fixtures. | Candidate truth; execution identity; review; semantic decision runtime; frozen policy. | Preserve deterministic retry/budget/state controls; route every repaired candidate through S4 recompilation and S6 acceptance before delivery. | src/agents/recovery.ts, src/core/run.ts, src/candidates/repair.ts, src/agents/reviewLifecycle.ts, src/architecture/candidateAssurance.ts, src/architecture/acceptanceOracle.ts; tests/recovery.test.ts, tests/candidateRepairProtection.test.ts, tests/candidateAssembler.test.ts, tests/candidateAssurance.test.ts, tests/runRepairCandidateLifecycle.test.ts, tests/reviewRepairCandidateLifecycle.test.ts. | CONTRACT; UNIT; INTEGRATION; SYSTEM_DETERMINISTIC; REAL_PROVIDER pending | 2026-09-24 S6-focused repair/review set within 11 files / 124 tests; full non-system 157 files passed / 1 skipped (931 passed / 7 skipped); scenario/concurrency system tests 2 files / 17 passed; REAL_PROVIDER pending. |
| Supply-chain provenance | Packed candidate receives verified SBOM, SLSA/in-toto provenance, BuildIdentity, and policy-required Cosign evidence. | Version-2 provenance remains bound to the exact CandidateRevision, configured packed artifact, BuildIdentity, policy and SBOM. Managed S8 handoff and finalization consume the existing policy-required supply-chain gate after current AcceptanceOracle acceptance and before public effects; standalone SDD handoff is explicitly unsupported. No S7 semantics changed. | S7 deterministic gate accepted 2026-09-24; S8 integration and CLI-surface remediation accepted by Campaign Director 2026-09-25. | REAL_PROVIDER signing/certification is not established; deterministic fixtures do not certify provider behavior. | S1 CandidateRevision and BuildIdentity; exact packed artifact; frozen supply-chain policy; S6 AcceptanceOracle-before-effects. | Keep required S7 provenance before public delivery effects; S13 retains provider certification. | src/provenance/generate.ts, src/delivery/handoff.ts, src/delivery/finalize.ts; tests/provenanceS7.test.ts, tests/deliveryFinalize.test.ts, tests/cliS8Handoff.test.ts. | CONTRACT; UNIT; INTEGRATION; BUILD; SYSTEM_DETERMINISTIC; REAL_PROVIDER pending | S8 remediation evidence is recorded below. |
| Delivery and reconciliation | Acceptance-gated exact effects with durable intent, receipt, and safe reconciliation. | S8 controller-owned flow persists intents, gates exact actions, and receipts execution or safe reconciliation. paseo.workspace.create is LOCAL_RESOURCE_CREATION, excluded from configured delivery effects, but remains S3-fenced and reconciled conservatively. The default issue path uses an internal controller workspace before acceptance and performs public issue/branch/push/PR effects only after current AcceptanceOracle acceptance and policy-required provenance. Uncertain execution is inspected before retry; unknown Paseo identity remains HUMAN_REQUIRED. The normal `sdd handoff` command returns explicit unsupported status before project loading and cannot enter this flow without a managed operation. | S8 deterministic gate accepted by Campaign Director 2026-09-25. External/provider lanes remain pending. | No known deterministic gap in the supported managed delivery flow. Standalone SDD handoff is intentionally unsupported; real GitHub/Paseo effects and complete packed provider reconciliation are not certified by fixtures. | Acceptance; authority; candidate identity; supply-chain policy. | S9 is the next eligible slice after S8 acceptance; S13 owns provider certification. | src/security/actionPolicy.ts, src/security/toolActionGate.ts, src/security/gatedAction.ts, src/security/actionReconciliation.ts, src/issues/workflow.ts, src/delivery/handoff.ts, src/delivery/finalize.ts, src/core/run.ts, src/cli.ts; tests/security/actionPolicy.test.ts, tests/security/gatedAction.test.ts, tests/security/toolActionGate.test.ts, tests/security/actionReconciliation.test.ts, tests/deliveryFinalize.test.ts, tests/cliS8Handoff.test.ts, tests/system/aehScenarioMatrix.test.ts, tests/system/aehConcurrencyCampaign.test.ts. | CONTRACT; UNIT; INTEGRATION; BUILD; SYSTEM_DETERMINISTIC; REAL_PROVIDER/PACKED_E2E pending | S8 remediation evidence is recorded below. |
| Runtime supervision | Controller owns process/provider leases, renewal, takeover, release, and cleanup. | Managed runtime mutations reload the shared snapshot under a root-scoped lock, validate it, and atomically persist service/provider state. The production Paseo adapter wraps operation-owned SDK materialization and participant/Lead turns with durable provider lease acquire, renewal, exact session inspection/stop, and release; current operation/candidate/policy/revision/token/epoch and participant or Lead-generation bindings are checked. Each lifecycle call has a unique owner ID, so overlapping writes cannot share one writer identity. Unsupported operation-owned SDK fallback fails closed. The lease's configured model-provider label is distinct from its Paseo-managed session inspection boundary. Expired leases remain fenced until observed quiescence or explicit reconciliation. Cancellation rechecks its current fence and releases provider leases only after exact-session quiescence. Control Center heartbeats its service record and drains owned service state on shutdown. R12 moves RUNNING/config/base-ref/portfolio bootstrap inside the existing catch, so strict base-ref errors terminalize as FAILED. | PARTIAL | Production adapter wiring, deterministic lease faults, and packed cancellation/takeover/restart-cleanup now have evidence. No REAL_PROVIDER startup/turn/renewal/cleanup receipt has been produced; provider certification remains a separate open lane. | Authority; accepted S5 session/context identity; supported Paseo provider. | Independently certify a supported Paseo lifecycle with actual provider startup, session identity, turn, renewal, quiescence and cancellation receipts; keep current operation/candidate/policy/revision/epoch fencing. | src/runtime/providerLifecycle.ts, src/runtime/supervisorV2.ts, src/runtime/managed.ts, src/paseo/runtimeCore.ts, src/paseo/runtimeInitialTurn.ts, src/operations/controller.ts, src/operations/state.ts; tests/paseoProviderLeaseLifecycle.test.ts, tests/paseoInitialTurnBarrier.test.ts, tests/runtimeSupervisorV2.test.ts, tests/security/controllerFencing.test.ts, tests/operations.test.ts, tests/operationLifecycleRegression.test.ts, tests/operationStateDurability.test.ts, tests/packed/s9ProviderLifecycleCampaign.mjs. | UNIT; INTEGRATION; BUILD; SYSTEM_DETERMINISTIC; PACKED_E2E (deterministic provider fixture); REAL_PROVIDER pending | 2026-09-25 S9 adapter, 13-file/137-test, concurrency, build, and packed lifecycle verification; R12 regression 1/1, typecheck, and fresh packed boundary inspection |
| Product, Control Center, and human interaction | Semantic projection, usable typed decisions, visible blockers, and safe control commands. | The authenticated loopback/CSRF server supplies the paired Human actor. Durable `OperationRecordV2` drives operation, candidate, stage/blocker, participant/error, counter, pause, and control projections; typed choices appear only for a current RUNNING/HUMAN_REQUIRED WAITING continuation whose request, continuation, frozen policy, candidate, execution revision, and epoch match. The App shows evidence and blocker detail, submits only operation/request/choice plus optional reason, refreshes after acceptance, and exposes server-derived pause/resume/cancel controls; the controller consumes exactly-once PAUSE/RESUME controls and applies pause only after a drain receipt proves writer quiescence. Server current-binding, expiry, replay, malformed, and mismatch checks remain authoritative. | PARTIAL | The mandatory real browser journey passes 46/46 assertions through the supported public start path and the real controller: current-scope review, UI choice submission, exactly-once consumption, continuation revalidation/resume, a second suspension, replay/stale rejection, pause/resume, scoped cancellation, and refreshed projections. The provider conversation was scripted for the fixture; packed human-interaction provider certification remains open. | Authority; runtime; operation state model. | Keep the current-binding checks and server-derived controls; provider-backed packed human interaction remains for the certification lane. | src/control-center/, src/security/humanDecision.ts, src/operations/state.ts, src/entry.ts, ui/control-center/src/api.ts, ui/control-center/src/App.tsx; tests/controlCenterOperationProjection.test.ts, tests/controlCenter.test.ts, tests/controlCenterApiContracts.test.ts, tests/humanDecision.test.ts, tests/operations.test.ts, tests/operationPauseResume.test.ts, tests/changeControllerOwnedDecision.test.ts, tests/browser/s9-control-center-journey.e2e.ts. | CONTRACT; UNIT; INTEGRATION; BUILD; BROWSER (46/46, scripted provider); PACKED human interaction pending | 2026-09-26 browser journey PASS; independent slice review pending |
| Observability and Engineering Evals | Candidate/execution-bound traces and metrics plus versioned reproducible advisory evals. | S12 adds the canonical `TelemetryCorrelationV1` identity (operation/candidate/execution/policy/controller/participant/runtime/session) with a domain-separated correlation digest across NDJSON events, OTel spans, and metric points; stale/wrong-candidate supplied identity is marked as an explicit violation instead of being silently re-attributed, and metric recording rejects a stale caller-supplied correlation. Real OpenTelemetry counters/histograms cover operation, participant, runtime-session, and validation signals and export canonical local snapshots with no credential. The versioned `evals/corpus/` manifest, cases, fixtures, deterministic scenario harnesses (bound into the corpus/case digest), and recorded build identity ship with recorded corpus/case digests and reproducible scoring. The advisory-invariant guard proves eval/telemetry output is never consumed by policy, authority, acceptance, permissions, or routing. | INDEPENDENTLY ACCEPTED (S12 R2, 2026-09-26). | No known deterministic S12 gap after the R1 repair. The OTLP lane is verified against an in-process loopback receiver only; hosted collectors and REAL_PROVIDER remain separate lanes. | Execution identity; certification. | No pending S12 acceptance gate. Keep evals advisory; describe the OTLP evidence as local loopback and treat hosted collector/REAL_PROVIDER evidence as separate future lanes. | [S12 WorkGraph](CORE_ARCHITECTURE_V2_S12_WORKGRAPH.md), Independent acceptance (S12-R2); [campaign evidence](evidence/s12/s12-observability-evals-campaign.json). | CONTRACT; UNIT; INTEGRATION; SYSTEM_DETERMINISTIC; LOCAL_OTLP loopback; REAL_PROVIDER pending | S12 R2 independently accepted 2026-09-26; campaign status PASS. |
| Security policy, isolation, sandbox, and SAST | OPA, role ceilings, tool gates, rootless isolation, network/scope controls, validator isolation, and candidate-bound SAST evidence. | S10 adds a deterministic rootless isolation provider: bwrap capability detection, minimal read-only host root, explicit toolchain binds, workspace-only writes, user/mount/PID/UTS/IPC namespaces, network denial by default, `--clearenv` plus environment allowlist, no-new-privileges, and explicit `ISOLATION_PROVIDER_UNAVAILABLE`/`ISOLATION_PROVIDER_UNSUPPORTED` blockers. Validator and external-tool commands run through it when `security.isolation.required`; the repository root is read-only except the declared writable paths (the workspace for configured validation commands and the evidence directory), as documented in SECURITY.md. Candidate-bound `SastEvidenceV1` binds adapter, tool version, command digest, raw-output digest, findings, workspace source digest, isolation evidence, and the exact CandidateRevision; stale, tampered, or missing evidence blocks, and the S4 security impact path requires it for opengrep/trivy. Real bwrap execution and a real local OSS Trivy v0.70.0 scan produced candidate-bound evidence for r1 (FAIL, secret finding) and r2 (PASS, clean). Podman/buildah are absent here and fail closed; OPA executable evaluation remains unavailable. Fresh independent S10 final review accepted this gate on reviewed head `87144f77a0c99088d385e848925643b5439422ea`; see the current re-acceptance record below.| ACCEPTED — S10 final independent review `SLICE_ACCEPTED` (2026-10-01).| No unresolved S10 acceptance gap. Live OPA evaluation and rootless Podman/OCI availability remain separate provider limits; required provider absence remains fail-closed.| S3 authority; S1 execution identity; local OSS isolation/SAST providers. | Preserve deterministic fail-closed isolation/SAST behavior; do not infer OPA or Podman/OCI certification.| [S10 final independent review](evidence/s10/review-s10-final.json); [clean candidate-bound Trivy evidence](evidence/s10/real-sast-campaign-85877349558477e21386d6e076393d15e2a90e26-release-1790866583369-387000-cc8194fb.json); hosted rootless job 110435254361 in run 36881830507.| CONTRACT; UNIT; INTEGRATION; BUILD; SYSTEM_DETERMINISTIC; ADVERSARIAL; REAL_PROVIDER (clean Trivy + hosted rootless bwrap); OPA_PROVIDER and rootless Podman/OCI not claimed.|S10 independently re-accepted 2026-10-01 on reviewed head 87144f7; hosted rootless run 36881830507 passed 3/3.|
| Contract and integration validation | Pact/OpenAPI/BDD and isolated service environments produce current candidate-bound reports and clean up. | S11 adds a candidate-bound `ProviderLaneEvidenceV1` module for CONTRACT/INTEGRATION/BROWSER/VISUAL lanes (candidate/workspace binding, provider identity, command/raw/artifact digests, stale/tamper/required rejection) and requires it for required contract/integration/BDD assertions in the S4 impact path. Declared specialized capabilities now resolve to matching providers instead of falling through to generic unit tests; missing Pact/OCI providers return explicit `CONTRACT_PROVIDER_UNAVAILABLE`/`INTEGRATION_PROVIDER_UNAVAILABLE` blockers, never a silent SKIP. The integration lifecycle requires explicit provision/readiness/test/cleanup steps, attempts cleanup after provision failure, and fails an absent readiness step. Real OpenAPI comparison and a real BDD runner produced candidate-bound CONTRACT evidence; a real loopback service lifecycle (provision → readiness → test → cleanup) produced candidate-bound INTEGRATION evidence with the service process asserted terminated and its pid file removed. The independent review rejection was repaired: a required specialized lane requirement can no longer be satisfied by a project script or raw command that merely exits zero; the S4 impact path loads and verifies the lane evidence persisted by the matching provider/adapter execution (OpenAPI now persists its own CONTRACT evidence) and fails closed with `PROVIDER_LANE_EVIDENCE_REQUIRED`. Cleanup blockers distinguish "no resources created" (`cleanupRequired:false`, no `INTEGRATION_CLEANUP_REQUIRED`) from "resources provisioned without a cleanup mechanism" (`INTEGRATION_CLEANUP_REQUIRED`). The R2 review then blocked on the declared `bdd`/`gherkin` route still bypassing the capability provider; it now routes through `runCapabilityValidator`, so a configured BDD provider command is honored, a missing runner returns typed `CONTRACT_PROVIDER_UNAVAILABLE` instead of the bare skip message, and a PASS persists candidate-bound CONTRACT evidence. The superseded `src/validators/gherkin.ts` adapter is deleted and the maturity inventory cites `src/validators/registry.ts`. | INDEPENDENTLY ACCEPTED (S11 R3, 2026-09-26; all 18 gate items PASS). | S11 contract/integration evidence is candidate-bound and provider-produced. Run 36824682741 on the last green implementation/workflow checkpoint before this documentation-only amendment also passed provider-contracts, validation provider contracts, real Pact provider verification, and the OCI integration environment contract. These CI checks do not claim live-model or broader packed UI certification. | Validation requirements; S10 isolation/sandbox; S9 browser harness. | No pending S11 acceptance gate. Preserve required-provider fail-closed behavior and keep any future browser/live-model/packed claims in their owning evidence lanes. | [S11 WorkGraph](CORE_ARCHITECTURE_V2_S11_WORKGRAPH.md), Independent acceptance (R3); [implementation-checkpoint run 36824682741](https://github.com/JamesMorales04/agentic-engineering-harness/actions/runs/36824682741). | CONTRACT; UNIT; INTEGRATION; SYSTEM_DETERMINISTIC; PACKED_E2E pending | S11 R3 accepted 2026-09-26; provider/Pact/OCI jobs passed on implementation checkpoint d21d6b5 in run 36824682741 (2026-10-01), before this documentation-only amendment. |
| Browser and visual validation | Browser/visual evidence is candidate-bound and traceable to packed UI behavior. | S11 adds a `visual-test` validation kind, a `visual` provider adapter, and a `UI/visual` impact dimension, plus candidate-bound BROWSER/VISUAL lane evidence with digest-bound screenshot artifacts. The pinned candidate `node_modules/.bin/playwright` is the only default provider (no bare npx fallback); a missing binary/browser fails closed with `BROWSER_PROVIDER_UNAVAILABLE`/`VISUAL_PROVIDER_UNAVAILABLE`, and a required browser/visual check without a current candidate binding fails with `PROVIDER_LANE_CANDIDATE_BINDING_REQUIRED`. The reusable S9 `tests/browser/` harness drives a real Control Center journey (pairing, rendered current-scope projection equal to durable operation/candidate/policy/revision/epoch/request identity, footer build identity, screenshots) and a visual journey (geometry, two-capture render stability, committed screenshot baseline) in real Chromium. Review remediation: each lane runs with its own durable Playwright output directory, so the VISUAL run can no longer delete the BROWSER screenshot; the campaign itself calls `verifyProviderLaneEvidenceV1` for both lanes after the full run and fails unless both verify with all artifacts present; and VISUAL evidence now binds the committed reference baseline digest plus the comparison configuration (`playwright-toHaveScreenshot`, `maxDiffPixelRatio: 0.05`), while a missing baseline fails with `VISUAL_REFERENCE_BASELINE_REQUIRED` and is never generated from the candidate. The sanitized machine summary is written inside the configured evidence output directory (`.harness/evidence/s11/browser-visual-campaign.json`), so the campaign no longer rewrites a source-tree file and its candidate/workspace digest recomputes MATCH after the run. The R2 LOW durability finding was also repaired: each lane's Playwright output directory is candidate-scoped, so a later campaign run against a different candidate cannot delete a prior candidate's screenshots (verified: the earlier `a20fd933b163` artifacts still verify after the newer `7ad7d70b2ef1` run). | INDEPENDENTLY ACCEPTED (S11 R3, 2026-09-26; all 18 gate items PASS). | The accepted S11 browser/visual evidence used real Playwright Chromium and the file-scripted deterministic Paseo boundary. Live-model/provider and broader packed browser/visual evidence are not claimed by that campaign and remain separate future evidence boundaries, not an outstanding S11 acceptance gate. | Validation resolver; CandidateImpact; Control Center; packed candidate. | No pending S11 acceptance gate. Preserve candidate-bound browser/visual evidence and certify additional live-provider or packed UI scope only through its owning lane. | [S11 WorkGraph](CORE_ARCHITECTURE_V2_S11_WORKGRAPH.md), Independent acceptance (R3); current hosted human journey is recorded in [CI run 36824682741](https://github.com/JamesMorales04/agentic-engineering-harness/actions/runs/36824682741). | CONTRACT; UNIT; BROWSER; VISUAL; PACKED_E2E pending; REAL_PROVIDER pending | S11 R3 accepted 2026-09-26; run 36824682741 confirms the separate human-e2e lane on implementation checkpoint d21d6b5 (2026-10-01), before this documentation-only amendment. |
| Real-provider certification matrix | Deterministic and real-provider lanes remain distinct; claimed capabilities have complete fresh packed-candidate evidence. | CertificationCore, 22-capability matrix, packed bootstrap, and Codex adapter exist. S7 verifies deterministic candidate/build/artifact/policy/builder/SBOM/signature bindings; its fake Cosign boundary is expressly fixture evidence and does not promote CertificationCore or the matrix. S13 repaired the packed bootstrap (`npm pack` prepare-noise parsing; packed artifact build identity bound into `CandidateRevision.metadata`), added a capability-journey oracle that emits each capability's `requiredEvidence` plus an oracle-verified `model.evidence` check, and executed a bounded real-provider campaign twice. Round 2 added product repairs for real Paseo completion: provider-appropriate semantic deadline (300 s), reasoning-class thinking bound, native-wait canonical-snapshot polling, supported Paseo timeline direction, typed-payload extraction, and the assessment JSON Schema in the launch prompt; the default `workhorse` alias was migrated to the available catalog id `mimo-v2.6-flash`. All four executed real-provider lanes are now ACCEPTED: `startup`, `informational` (real packed `intent` through a real Paseo `opencode-go/mimo-v2.6-flash` session with thinking `low` and a durable INTENT assessment cache), `project-home`, and `multi-project`, each with a fresh packed artifact, packed build identity, real session receipt, and deterministic oracle. | INDEPENDENTLY ACCEPTED (S13 final Round-23 review, 2026-09-30; complete 22-row F2-final matrix). | The final review records 21 `F2_CURRENT_ACCEPTED` rows and `control-center` as the single `NOT_APPLICABLE` row; the three `paseoSdkResolve` mise-drift tests and the disclosed fixture-level model override are explicit boundaries, not failed S13 matrix rows. AEH-V2-0132/0133/0134 and test-fixture hygiene are carried as nonblocking post-v2 observations after S14/closure acceptance. | All claimed capability paths; acceptance and evidence gates; S13 real-provider certification. | No pending S13 acceptance gate. Preserve the accepted 22-row matrix and its N/A/observational boundaries; post-v2 observations remain tracked separately. | [S13 final independent review](evidence/s13/review-final-r23.json) (`SLICE_ACCEPTED`); [S13 WorkGraph final acceptance](CORE_ARCHITECTURE_V2_S13_WORKGRAPH.md); [closure review](evidence/closure/review-closure.json). | CONTRACT; UNIT; BUILD; REAL_PROVIDER (4 accepted rows); PACKED_E2E (4 accepted rows); BROWSER/ADVERSARIAL separately reported; REAL_PROVIDER/PACKED_E2E incomplete + SLICE_BLOCKED for 18 rows | S13 final Round-23 review accepted 2026-09-30 on F2-final buildDigest `599ed363…`; closure review accepted 2026-09-30. |
| Learning and feedback | Versioned post-run candidates; independent evaluation and governed adoption; no implicit authority mutation. | Memory recall, telemetry, evals, and within-run repair exist; no cross-run learning pipeline. | DEFERRED | Learning and persistent skill promotion are explicitly post-v2. | Observability; human-governed policy. | Reopen only with versioned candidate and explicit adoption gate. | src/memory/, src/evals/; advisory use only. | UNIT for advisory contracts; adoption lane deferred | 2026-09-23 audit snapshot; not rerun |
| Self-hosting readiness | Deterministic composite gate plus disposable self-modification fixture. | S14 composite policy is deterministic and `modelAuthority: NONE`; after closure it is `ready: true` with 21/21 required items, 47 verified artifacts, 0 failed, policy digest `1be97df3…`. The closure resource-stability campaign returned two real governed operations to baseline 0/0/0 using product reconciliation (no campaign-side cleanup); the disposable campaign completed 13/13 receipt dispositions with checkout bytes unchanged. | ACCEPTED — S14, final closure, and S10 are independently accepted; Core Architecture v2 is complete for its documented scope.| Unrelated nonblocking post-v2 observations remain open: AEH-V2-0132/0133/0134, `tests/paseoSdkResolve` environment drift, and S14-F4 fixture hygiene. Closure provider flakes remain historical failed-run evidence.| All certified S1–S14 gates and the closure evidence. | Preserve accepted S1–S14/closure evidence and keep post-v2 observations separate; next product activity is the human-driven cold-start trial after release.| [S14 independent review](evidence/s14/review-s14.json) (`SLICE_ACCEPTED`, no blocking action); [closure independent review](evidence/closure/review-closure.json) (`SLICE_ACCEPTED`); [closure WorkGraph](CORE_ARCHITECTURE_V2_CLOSURE_WORKGRAPH.md); [closure resource stability](evidence/closure/resource-stability-product.json); [closure composite](evidence/closure/self-hosting-composite-report.json). | DETERMINISTIC; PACKED_E2E; REAL_PROVIDER versioned; ADVERSARIAL |S14 and closure accepted 2026-09-30; S10 accepted 2026-10-01; hosted follow-up CI run 36881830507 passed 13/13.|

**S13 round-12 advance (2026-09-28, implementing lead; `SLICE_BLOCKED`; independent review pending):** round 12 repaired the P0-A Spec Manager canonicality defect (AEH-V2-0115: typed pre-persistence `SPEC_MANAGER_CONTENT_NOT_CANONICAL` gate plus a canonical per-capability `{ capability, content }` delta contract and per-capability `specs/<capability>/spec.md` persistence) and isolated/repaired the P0-B DELEGATED escalation constraint (AEH-V2-0110: escalation now requires the Planner's typed `formalizationNeed: REQUIRED`, with the constraint, exact Planner judgment and contributing Explorer findings recorded in `harness.change.triage-escalated`). The live `change-delegated` lane then exposed and round 12 repaired seven further real pipeline defects (contract-before-planning with frozen requirement ids; escalation entering the formalization path in the same operation; provisional bootstrap policy superseded exactly once by the first execution-semantics bind; known technology competencies selecting their defining seed skill; explicit launch-permission projection from the compiled tool ceiling; single-approved-command resolution for `command`-kind requirements; approval-stop turns failing instead of silently succeeding), each with focused regressions. `harnessRevisions` is now bound into every per-lane envelope (R11-F5), authorization history is preserved with a mandatory issue-create confirmation (R11-F1-residual), and the post-run tracked-doc edit window is recorded (R11-F6). The lane reaches planning, participant compilation, a real Implementer session and a PASS validation report, then blocks on **AEH-V2-0116**: a Paseo-managed OpenCode Implementer session stops at the provider approval prompt (`permission`) despite an explicit permission projection and `featureValues.auto_accept: true`, so no ChangeSet is produced. `delegated-change` and `formal-sdd` remain NOT_ACCEPTED; the live canonical FORMAL_SDD rerun and the P2-F matrix rows (multi-worker, repair family, context-handoff, permission-delegation, distributed-execution, full issue-driven, delivery) were not reached and hand off to Round 13. Round-12 validation: focused 20 files / 172 tests passed; full non-system 200 files (198 passed, 1 skipped; the 3 `paseoSdkResolve` failures are the documented pre-existing mise/Paseo-CLI environment drift); `npm run typecheck` and `npm run build` exit 0; `git diff --check` clean; TARGET byte-identical (`4954b62b…`). No row or slice is self-accepted; S14 must not start.

**S13 round-13 advance (2026-09-28, implementing lead; `SLICE_BLOCKED`; independent review pending):** round 13 diagnosed AEH-V2-0116 at the Paseo daemon boundary and repaired all AEH-owned parts: the OpenCode projection now declares a scoped `external_directory` allow for the frozen launch root and its git metadata roots (D1, `tests/permissions.test.ts`, `tests/paseoLaunchSpec.test.ts`); the pending approval payload returned by `waitForFinish` is captured into a bounded failed-turn detail instead of being discarded (D2, `tests/paseoRuntime.test.ts`); Codex participants receive a provider-native sandbox projection `approval_policy: never` + `sandbox_mode` from the compiled ceiling instead of the provider-default `on-request` policy (D3); and isolated launches no longer carry the Paseo parent handle because Paseo child agents inherit the parent workspace/cwd and were executing isolated wave tasks inside the operation candidate workspace (D4 / AEH-V2-0117, probe6). Two provider-specific behaviors are documented rather than waived: Paseo deterministically disables `auto_accept` whenever a `toolPolicy` is present (`opencode-agent.js:2425`), and child agents inherit the parent workspace/cwd. The repaired real lane `run-20260928225446983-c5ff6c` then reached Implementer COMPLETED with a validated ChangeSet and assembly receipt r2->r3, and exposed the next deterministic blocker **AEH-V2-0118**: DELEGATED remediation launched repairers after a PASS validation report (including `contract-test` PASS), both repairers returned no-change, and the operation terminalized FAILED at revision 122 with no durable error. The campaign harness now persists per-stage messages and failed-participant artifacts (R12-F4). `delegated-change` remains NOT_ACCEPTED; `formal-sdd`/full issue-driven depend on the same trunk. Evidence: `docs/evidence/s13/round13-0116-diagnosis.json`, `docs/evidence/s13/round13-run3-cleanup-annotation.json` (R12-F3), lane `docs/evidence/s13/governed-lanes/round-13/run-20260928225446983-c5ff6c/change-delegated.json`. The R12-F1 post-Round-13 build-bound sweep passed for `startup`, `informational`, `project-home`, `multi-project` (certification summary sha256 `9b4efcb0…`, packed build `release-1790636944026-740060-fa9fa05c`, `checkoutUntouched: true`) and for governed `audit` (operation `AUDIT-20260928T233544Z-78e1af93` terminal SUCCEEDED, lane sha256 `7fe095fb…`); the `change`/`direct-change` rerun and the GitHub `issue-driven` rerun are the remaining sweep handoffs. R12-F3 is annotated `LEAD_OBSERVED_DAEMON_CORROBORATED`; the campaign harness now persists per-stage messages and failed-participant artifacts (R12-F4); the R12-F5/F6 statements are corrected. No row or slice is self-accepted; **S13 round-15 advance (2026-09-29, implementing lead; `SLICE_BLOCKED`; independent review pending):** round 15 reconciled the unreviewed Round-14 edits (all ten source files completed and covered by focused tests; nothing reverted) and completed the reviewer provider-stop ownership repair: **AEH-V2-0119** — a Reviewer provider/runtime stop (non-zero provider session, unparseable contract, or candidate mutation attempt) now terminalizes immediately with a durable typed `SYSTEM_FAILURE` check naming the reviewer, kind and session detail instead of looping through no-change Repairers as remediation debt; the Round-14 remediation budget remains as a deterministic bound for genuine semantic debt, the reviewer prompt states its authorized read boundary and the changed files/requirements, and the control-root `external_directory` read projection is retained (`tests/reviewRepairCandidateLifecycle.test.ts`, `tests/reviewLifecycle.test.ts`, `tests/paseoLaunchSpec.test.ts`). **AEH-V2-0120** — supervisor generations rotate when their structured-result binding belongs to a superseded candidate (fixing `AEH_RESULT_STALE_CANDIDATE` continuations), and every terminal review result carries a quality state so the run summary no longer crashes (the Round-15 run-2 `debtScore` TypeError); `tests/supervisorConsolidationProvenance.test.ts`, `tests/reviewRepairCandidateLifecycle.test.ts`. **AEH-V2-0121** — the governed fixture now provides a real `check:architecture` validation for the canonical architecture impact dimension, and the campaign binds a real Paseo Lead agent (`aeh.provider` label) per lane through `PASEO_AGENT_ID` so headless DELEGATED operations can produce the Lead semantic evidence the AcceptanceOracle requires; no oracle/validator gate was weakened. Three packed `change-delegated` lanes on the current source then advanced from architecture-assurance blocking (run 1), through the reviewer-stop terminal path (run 2, whose summary crash was fixed), to `review=ACCEPTED` (run 3), which now fails closed only at `ACCEPTANCE_LEAD_REQUIRED` (headless Lead binding, now wired). A second machine restart (~2026-09-29T02:46Z) interrupted lane attempt 4 and cleared its fixture; the interrupted Round-13/14 operations remain unobservable and no terminal state is claimed. Evidence: `docs/evidence/s13/round15-environment-loss.json`, `docs/evidence/s13/governed-lanes/round-15/run-20260929-r15-delegated-{1,2,3}/change-delegated.json`, summaries `s13-governed-operations-round-15-run-*.json`. `delegated-change` remains NOT_ACCEPTED; `change`/`direct-change`, `issue-driven`, canonical FORMAL_SDD and the P2-F matrix reruns continue. S13 stays uncommitted; S14 must not start.

S14 must not start.

**S13 round-15 final lanes (2026-09-29):** the packed `change-direct` lane reached a real terminal `SUCCEEDED` on the current source (AcceptanceOracle `ACCEPTED`, 0 blockers, rows `change`+`direct-change` PASS, 3 real Paseo sessions, `checkoutUntouched: true`, 4 workspaces archived / 0 remaining; lane sha256 `4865e915b344b4eab5b54e91c284960683d81316eb150886604bbbf8965948a6`), and the GitHub campaign under the external user authorization (`docs/evidence/s13/github-effect-authorization.json`) with the user-authorized existing issue #1 passed `issue-driven` (snapshot/`ISSUE_DRIFT` gate PASS, operation-bound import completed) and `authority` (unapproved `HUMAN_REQUIRED`, approved `EXECUTED/SUCCEEDED`, real branch `aeh-s13-authority-20260929043030`) with no issue/repo created, edited or deleted; overall `SLICE_BLOCKED` on `delivery`. The FORMAL_SDD live lane now passes spec-authoring, canonical persistence and OpenSpec compilation (with the new pre-persistence SHALL/MUST normative gate, AEH-V2-0111-residual, and the explicit `AEH_RESULT_JSON=` marker instruction, AEH-V2-0125) and is blocked only by **AEH-V2-0126** (OPEN): authoring persistence into the candidate workspace changes its digest without a deterministic candidate advancement, so the Planner result is rejected `CANDIDATE_WORKSPACE_MISMATCH` (run 3 sha256 `6f134a4e7346df06…`). Round-15 final validation: 18 focused files / 152 tests green, `npm run typecheck` and `npm run build` exit 0, `git diff --check` clean, TARGET byte-identical, all lane workspaces archived, no credential persisted.

**S13 round-17 advance (2026-09-29, implementing lead; `SLICE_BLOCKED`; independent review pending):** round 17 reconstructed the interrupted Round-16 record from durable artifacts (final freeze `2bee2ec0…`; `r16-formal-7` reached a product terminal SUCCEEDED but the campaign oracle misread the durable stage object as a string), then verified the campaign stage-shape fix with `change-formal` on that freeze (`r17-formal-1`: terminal SUCCEEDED, oracle PASS). It implemented the missing `delivery` capability: `delivery.github.pullRequests` (default true) enumerates the requested effect set in frozen policy, so an accepted candidate can be finalized push-only within the user's external authorization (which covers `git.push`/`github.branch.create`/`github.issue.create` but not `github.pull-request.create`). The new real lane drives accepted candidate → authorized delivery intent → `HUMAN_REQUIRED` deny → external-authorization-derived APPROVE bound to the exact effect digest → branch create + commit + push → `FINALIZED` → remote ref observed, with 3 durable intents/receipts, 0 pull requests and issue #1 untouched. The product change forced a re-freeze: trackedDigest `5d1b6680ee1d4f6b7ff623521ceedb77ae70e3f4180d2a48946e3e1c7adc4d7e` (668 files), build `release-1790684324214-110138-2fd31d7e`, and every current row was re-run on it: `change-formal` (`r17-formal-3`), `change`/`direct-change` (`r17-direct-1`), `audit` (`r17-audit-2`), `delegated-change` (`r17-delegated-1`), certification 4 (`r17-certification-2`), `cancel`/`recovery` (`r17-lifecycle-2`) and GitHub `issue-driven`/`authority`/`delivery` (`r17-gh-4`) — all PASS with real providers, `checkoutUntouched: true`, workspaces archived. AEH-V2-0110/0111/0115/0126 are closed by the final-freeze formal/delegated lanes and AEH-V2-0127 records the delivery effect-enumeration repair. Remaining: full issue-driven end-to-end (the authorized issue #1 is not implementation-grade and the authorized repository has no configured validation), the P1-E matrix rows (`multi-worker`, repair family, `context-handoff`, `permission-delegation`, `distributed-execution` — `BLOCKED_NO_EVIDENCE`, no harness fabricated), and the full suite's 3 pre-existing `paseoSdkResolve` failures. Round-17 validation: focused 24 files / 205 tests; canonical non-system suite 1,347 passed / 3 failed (pre-existing) / 7 skipped; `npm run typecheck`, `npm run build` and `git diff --check` clean; TARGET byte-identical (`4954b62b…`). No row or slice is self-accepted; S14 must not start.

## Approved dependency DAG

The DAG reflects the reconciled S1-S14 ownership. Existing candidate truth and controller-fencing foundations are prerequisites. S2, S3, and S5 can proceed in parallel after S1 interfaces are frozen. S7 may progress alongside S4-S6; S9 may proceed alongside S6-S8 after authority and context identity contracts are fixed. S10 is the security isolation slice. S11 adapters may be developed earlier, but their certification evidence waits for stable acceptance and policy interfaces.

~~~mermaid
flowchart TD
  F[Existing candidate truth and controller-fencing foundation] --> S1[S1 Frozen Execution and Skill Identity]
  S1 --> S2[S2 Production Semantic Decision Runtime]
  S1 --> S3[S3 Authority Kernel and Human Control]
  S1 --> S5[S5 Progressive Context and Paseo Session Identity]
  S2 --> S4[S4 CandidateImpact and Assurance Recompilation]
  S3 --> S4
  S5 --> S4
  S3 --> S6[S6 AcceptanceOracle and Objective Completion]
  S4 --> S6
  S1 --> S7[S7 Supply-Chain Provenance Gate]
  S6 --> S8[S8 Controller-Owned Delivery and Reconciliation]
  S7 --> S8
  S3 --> S8
  S3 --> S9[S9 Runtime Supervision and Control Center]
  S5 --> S9
  S3 --> S10[S10 Security Isolation, Sandbox and SAST]
  S1 --> S10
  S6 --> S11[S11 Contract, Integration, Browser and Visual Validation]
  S9 --> S11
  S10 --> S11
  S1 --> S12[S12 Observability and Engineering Evals]
  S6 --> S13[S13 Real-Provider Certification Matrix]
  S7 --> S13
  S8 --> S13
  S9 --> S13
  S10 --> S13
  S11 --> S13
  S12 --> S13
  S13 --> S14[S14 Self-Hosting Composite Gate]
~~~

The cross-slice dependency summary is:

- S1 → S2, S3, S5; it also establishes identity inputs for S7, S10, and S12.
- S2 + S3 + S5 → S4.
- S3 + S4 → S6.
- S7 can progress alongside S4-S6 after candidate/build identity is available; S6 + S7 + S3 → S8.
- S3 + S5 → S9, which can proceed alongside S6-S8.
- S3 + S1 → S10; S6 + S9 + S10 → S11.
- S12 proceeds in parallel after S1.
- S13 requires the implemented and gated paths for each capability being claimed, including relevant S6-S12 evidence.
- S14 requires all required capability certifications and the disposable self-modification campaign.

## S1-S14 implementation slices

### S1 — Frozen Execution & Skill Identity

**Objective:** establish truthful, immutable policy, role, skill, blueprint, binding, context/prompt, and structured-result identity at launch and result boundaries.

**Dependencies:** existing candidate truth and controller-fencing foundation.

**Definition of Done:**

1. A deterministic compiler produces immutable ResolvedOperationPolicy and per-role RoleInvocationPolicy; model output cannot choose, weaken, or widen either.
2. A versioned ExecutionBlueprint binds project, operation execution revision, candidate revision/digest, controller epoch, policy, WorkGraph/ParticipantPlan/catalog, participant role/specialization, ToolPack, resource claims, validation resolution, output contract, and SkillManifest digests.
3. Per-participant ExecutionBinding binds generation, approved runtime/model/session, actual ContextManifest and PromptManifest digests, and applicable controller-issued lease identity.
4. Accepted ephemeral procedure content is retained in SkillManifest, projected verbatim only to its assigned authorized participant, and included in the prompt/binding digest. An ID-only assignment fails.
5. Every production structured-result launch path propagates the full versioned binding. The gateway rejects missing, stale, replayed, wrong-candidate, wrong-generation, wrong-epoch, wrong-policy, wrong-blueprint, wrong-skill/context/prompt, wrong-output-contract, and wrong-runtime results.
6. Candidate assembly invalidates all execution bindings and results for the previous candidate. Controller takeover invalidates old-epoch writers and leases.
7. Obsolete identity versions fail with explicit unsupported-version/migration errors; no dual-read compatibility path is added.
8. Deterministic tests cover valid propagation and each rejection case across direct, Podman, Paseo, and distributed transports. These tests do not claim real-provider certification.

recordRevision remains the append-only event/order revision. operationExecutionRevision changes only when execution semantics change, so routine stage events do not stale active bindings.

**Gate:** all eight Definition of Done items hold across production launch/result paths; focused deterministic evidence is complete. No real-provider certification is implied.

### S2 — Production Semantic Decision Runtime

**Objective:** wire bounded semantic assessment into public intent, route, stack, issue, failure, and impact paths, with typed evidence and deterministic post-checks.

**Dependencies:** S1 identity interfaces and the approved model/provider interface.

**Gate:** each public semantic path has an explicit DETERMINISTIC, MODEL, or HYBRID mechanism, preserves unknowns and provenance, and cannot grant authority or bypass policy.

### S3 — Authority Kernel & Human Control

**Objective:** consume scoped HumanDecisions; enforce controller/current-epoch ownership and cancellation fencing; correct OPA canonical-role matching; and make action and policy authority semantics consistent across mutations and effects.

**Dependencies:** S1 execution identity and existing candidate truth.

**Gate:** each decision, mutation, cancellation, and sensitive action is checked against current operation, candidate, policy, actor, scope, lease, and epoch; product choice remains distinct from action authorization.

**Ownership boundary:** security-policy facts required for role ceilings and action authority belong here. Rootless sandbox, network isolation, validator isolation, provider sandbox certification, and SAST closure belong to S10.

### S4 — CandidateImpact & Assurance Recompilation

**Objective:** bind actual post-assembly impact to the current candidate and recompile assurance, reviewers, validation, acceptance assertions, and evidence strength from that impact and policy.

**Dependencies:** S2 semantic runtime, S3 authority/policy, S5 context/session identity, and candidate truth.

**Gate:** no impact result can weaken a policy floor; required dimensions produce required independent evidence; missing or unresolved high-risk impact blocks acceptance.

### S5 — Progressive Context & Paseo Session Identity

**Objective:** wire full progressive Context Runtime v2, JIT/addressable references, continuation semantics, retrieval lifecycle, broader progressive projection, and create/reuse/rotation of PaseoSessionBinding.

**Dependencies:** S1 execution identity and a frozen controller-issued context-authorization interface. S5 can proceed alongside S3 after the shared interface is fixed; S3 owns authority semantics.

**Gate:** each launch/result binds the actual authorized context and prompt; continuation and retrieval remain candidate/session-bound; any session binding mismatch rotates or rejects reuse.

### S6 — AcceptanceOracle & Objective Completion

**Objective:** complete AcceptanceAssertion, VerificationRequirement, OracleResolver, EvidenceBundle, and AcceptanceOracle disposition; enforce objective Definition of Done and managed Lead acceptance ordering before delivery.

**Dependencies:** S3 authority/human controls and S4 impact-driven assurance.

**Gate:** every required assertion has sufficient current candidate-bound evidence; validation PASS or reviewer opinion alone cannot accept; completion evaluates the complete objective evidence set.

**Campaign status:** S6 and S7 were accepted on their deterministic gates, 2026-09-24. S8 was independently accepted by the Campaign Director on its deterministic gate on 2026-09-25.

### S7 — Supply-Chain Provenance Gate

**Objective:** bind packed-candidate SBOM, provenance, BuildIdentity, and policy-required signing evidence to the artifact and make required provenance a deterministic gate.

**Dependencies:** candidate/build identity. This slice may progress alongside S4-S6.

**Gate:** missing, tampered, stale, or wrong-builder evidence blocks the delivery path wherever policy requires it. Deterministic fixtures remain distinct from provider-backed signing evidence.

**Campaign status:** the deterministic S7 gate, including the schema-guard remediation, was accepted by the Campaign Director on 2026-09-24. S8 was independently accepted by the Campaign Director on its deterministic gate on 2026-09-25.

### S8 — Controller-Owned Delivery & Reconciliation

**Objective:** route every requested external effect through controller-owned ActionIntent, ToolActionGate, ActionReceipt, and safe reconciliation.

**Dependencies:** S3 authority, S6 acceptance, and S7 provenance where required.

**Gate:** no effect occurs before current acceptance and required policy gates; uncertain non-idempotent outcomes are inspected and reconciled before retry or completion.

**Campaign status:** **SLICE_ACCEPTED** by the Campaign Director on 2026-09-25 after independent diff review and deterministic reruns. The standalone `aeh sdd handoff` command is explicitly unsupported because no managed accepted-delivery lifecycle is available to a normal CLI invocation. S9 had not started at the time of acceptance.

**S8 remediation evidence — 2026-09-25:** The final focused S8 set passed 14 files / 198 tests. `npx tsc --noEmit` and `npm run build` exited 0 (release `release-1790322187860-3232428-0507f39b`). Full non-system Vitest passed 162 files with 1 skipped (1,014 passed, 7 skipped); affected deterministic scenario/concurrency suites passed 2 files / 17 tests. The new CLI regression passed 2 tests and proves unsupported handling occurs before project-config loading and produces no workspace/public effects. The Flash worker's baseline run failed both tests against the stale command and passed both against the remediation. The Campaign Director independently reran the focused S8 set (7 files / 89 tests), TypeScript, build, full non-system suite, and affected system suites; all passed. `git diff --check` and the normative-target equality check passed. Fixtures are deterministic evidence; REAL_PROVIDER, BROWSER, OPA_PROVIDER, ADVERSARIAL, and PACKED_E2E claims remain pending. **Campaign Director disposition: SLICE_ACCEPTED.**

### S9 — Runtime Supervision & Control Center

**Objective:** complete controller-owned process/provider lifecycle, leases, renewal/release, takeover, drain/cleanup, semantic operation and participant state projections, and usable typed Control Center decisions/controls.

**Dependencies:** S3 authority and S5 session/context identity. This slice may proceed alongside S6-S8.

**Gate:** single-writer ownership, current-epoch fencing, safe cancellation/resume, accurate projections, and scoped decisions are evidenced end to end.

**Campaign status:** **IMPLEMENTATION_COMPLETE — pending independent slice review.** The mandatory real browser journey passed against the current candidate (1 Playwright test, 46/46 assertions) through the supported public start path, a real Playwright Chromium, the real Control Center UI/server, and the real controller: pairing, rendered current-scope review, UI decision submission, exactly-once controller consumption, continuation revalidation/resume, replay/stale rejection, pause with durable `PAUSED` after drain/fencing, resume, scoped cancellation with writer fencing, and refreshed API/rendered projection equality. The external model conversation was replaced by the file-scripted deterministic Paseo boundary (`AEH_DETERMINISTIC_PASEO_RUNTIME=1`); that is fixture evidence, not REAL_PROVIDER certification. The implementing lead does not accept its own slice; a fresh independent reviewer owns slice acceptance. S10 remains locked.

**R12 remediation evidence — 2026-09-25:** the packed Lead runtime's selected release had config/base-ref/portfolio bootstrap outside the controller's `try`, so strict `baseRef=main` failure in a non-Git disposable root exited after persisting `RUNNING` and left a dead-owner operation nonterminal. The repair in `src/operations/controller.ts` places that bootstrap inside the existing fenced failure-finalization boundary; strict base-ref failure still terminalizes as `FAILED`. The delegated Flash test reproduced the old code: `npx vitest run tests/operationStartupFailure.test.ts` in isolated baseline `c036471` exited 1 with 1 test failed at `resolveBaseRef` before the catch. After selective integration, `./node_modules/.bin/vitest run tests/operationStartupFailure.test.ts --reporter=dot` passed 1 file / 1 test and `npm run typecheck` exited 0. The Campaign Director independently ran the earlier test draft — 1 file / 1 test passed; typecheck exited 0; build exited 0 (`release-1790359341925-17-d65e1b0f`). The fresh packed R12 artifact is `/tmp/aeh-s9-r12-pack/agentic-engineering-harness-0.8.4.tgz`, SHA-256 `77a0c39f580053b67abe2ee444e6b4c2995ecdc417a66613eb56d5dc20a1782d`, selecting `release-1790359422074-34-41c17389`; packed inspection confirms the bootstrap is inside the catch. The Director owns initialization/recovery of the same disposable operation; no S9 browser decision or cancellation has been submitted by the implementation lead.

**Browser journey evidence — 2026-09-26:** the maintained Playwright harness in `tests/browser/` now starts a real governed CHANGE operation through the public `npm run aeh -- operation start change` path with the file-scripted deterministic Paseo runtime, so the real controller owns the Spec Manager suspension, exactly-once consumption, and resume. The journey passed 46/46 assertions against candidate release `release-1790386441389-339964-cb4c88d0` (gitSha `c036471`, dirty, package 0.8.4): pairing, rendered current scope (operation, candidate digest, policy digest, `operationExecutionRevision`, controller epoch), UI decision submission, exactly-once consumption with replay/unknown/out-of-scope/foreign/CSRF rejection, continuation revalidation and resume, a second controller-owned suspension, pause to durable `PAUSED` with an empty drain receipt and stale-writer fencing, resume back to `HUMAN_REQUIRED`, scoped cancellation to `CANCELLED`, and final API/rendered projection equality. Sanitized evidence: `docs/evidence/s9/s9-playwright-journey.json`, `s9-playwright-paired-home.png`, `s9-playwright-final-projection.png`. This is BROWSER plus deterministic-provider fixture evidence, not REAL_PROVIDER certification.

### S10 — Security Isolation, Sandbox & SAST

**Objective:** prove rootless sandbox and network/scope isolation, validator-command isolation, and candidate-bound SAST evidence under the authority policy.

**Dependencies:** S1 execution identity and S3 role/action policy.

**Gate:** prohibited actions and out-of-scope access are denied; sandbox/provider evidence and SAST reports are candidate-bound. This provider/isolation certification belongs to S10 even when policy semantics are prerequisites in S3.

**Historical campaign status (before final re-acceptance):** **IMPLEMENTATION_COMPLETE — pending independent slice review.** The later fresh acceptance is recorded below. The S10 WorkGraph, frozen provider/evidence contracts, delegation assessment, and exact environment facts are in [S10 WorkGraph](CORE_ARCHITECTURE_V2_S10_WORKGRAPH.md). The pre-S10 gaps were independently reproduced by a bounded worker in an isolated Paseo worktree: [baseline gaps](evidence/s10/baseline-gaps.md).

**Environment facts:** kernel `6.18.52-1-cachyos-lts`, uid 1000, `kernel.unprivileged_userns_clone = 1`; `bwrap` (bubblewrap) `0.13.0` available; `runc`, `unshare`, `newuidmap`, `newgidmap` available; `podman`, `buildah`, `crun`, `slirp4netns`, `fuse-overlayfs`, `opa`, `opengrep`, `semgrep` absent; `sudo` unavailable (password required), so no system package installation. Trivy `v0.70.0` (Apache-2.0) was provisioned as a user-local OSS binary with the published release checksum (`trivy_0.70.0_Linux-64bit.tar.gz` SHA-256 `8b4376d5d6befe5c24d503f10ff136d9e0c49f9127a4279fd110b727929a5aa9`; extracted binary SHA-256 `379d59f24a4a828c55de5f0b91b6805cc35d13580180b658820e648611256166`) for real SAST execution only; it is not a committed dependency.

**Real rootless isolation evidence — 2026-09-25:** `tests/system/aehIsolationCampaign.test.ts` executes the actual `bwrap` provider (3 tests passed) and proves, from inside the sandbox: the host uid is preserved without privilege (`uid=1000`), `CapEff=0000000000000000` and `NoNewPrivs=1`; only `lo` exists in the new network namespace and a real TCP connect to `1.1.1.1:80` fails with `ENETUNREACH`; a live host process PID is not visible under `/proc`; host `$HOME/.ssh` and sibling host directories are masked; `/etc` is read-only; writes to the declared workspace land on the host while writes outside it do not; an inherited host environment secret is absent. The validator-isolation case runs a required opengrep-adapter command inside the sandbox: repository mutation is denied, egress is denied, the check passes on valid structured output, and the check records full isolation evidence. The focused deterministic contract suites `tests/security/isolation.test.ts`, `tests/security/validatorIsolationFailClosed.test.ts`, and `tests/security/providerFailClosed.test.ts` cover argument construction, unsupported providers, environment rejection, and explicit `ISOLATION_PROVIDER_UNAVAILABLE`/`ISOLATION_PROVIDER_UNSUPPORTED` blockers; an unavailable provider never runs the validator and never reports a silent SKIP or PASS.

**Candidate-bound SAST evidence — 2026-09-25:** `S10_TRIVY_BIN=/tmp/opencode/s10-tools/trivy/0.70.0/trivy node tests/packed/s10RealSastCampaign.mjs "$(pwd)"` ran against built release `release-1790389747122-486790-a8c37684` and passed all assertions. Candidate r1 (planted GitHub PAT) produced a real FAIL with one `github-pat` finding and candidate-bound `SastEvidenceV1`; candidate r2 (secret removed) produced a clean PASS with zero findings. Both artifacts are persisted under the candidate-scoped directory `.harness/evidence/sast/<candidateId>-r<revision>-<identityDigest12>/`, bind the exact candidate/workspace source digest, tool version, command digest, raw-output digest, findings, and isolation evidence, and verify by digest. The r1 evidence is rejected as `SAST_EVIDENCE_STALE` for r2, proving exact-revision traceability. Deterministic contract tests (`tests/security/sastEvidence.test.ts`) cover persist/load/verify, stale/tampered/missing rejection, workspace drift refusal, and fail-closed missing-tool behavior. Sanitized summary: [real SAST campaign](evidence/s10/real-sast-campaign.json). This is REAL_PROVIDER evidence for a local OSS scanner plus a real rootless provider; it does not certify hosted providers, rootless Podman/OCI execution, or S13 provider certification.

**Deterministic/build evidence — 2026-09-25:** `npm run typecheck` exited 0; `npm run build` exited 0 (`release-1790389747122-486790-a8c37684`). The full non-system suite passed 175 files with 1 skipped (1,134 tests passed, 7 skipped) except `tests/paseoSdkResolve.test.ts` (3 failures); that file was rerun in a clean detached worktree at baseline `0afebcd` and failed 3/3 there as well, confirming environment drift from a mise-installed Paseo CLI rather than an S10 regression. The relevant system lanes passed: scenario matrix, concurrency campaign, and isolation campaign — 3 files / 20 tests. The adversarial suite passed 26 tests with its two long-recorded S6 objective-completion fixture failures unchanged. Rootless Podman/OCI execution, OPA executable evaluation, BROWSER, and complete PACKED_E2E certification remain separate open lanes owned by later slices.

**Historical independent-review remediation — 2026-09-26:** the fresh independent reviewer rejected the first S10 evidence because the isolation campaign's host-home probe tested `${os.homedir()}/Desarrollo`, which npm/npx turns into a bwrap bind-destination parent; the campaign therefore failed under `npm test`/`npx vitest` while passing with a direct binary invocation. The probe now asserts a host-only home sentinel that is independent of bind-destination parents, and the security row documents the configured-validator workspace exception consistently with SECURITY.md. Re-verification: `npm test -- tests/system/aehIsolationCampaign.test.ts` 1 file / 3 tests passed; `npx vitest run tests/system/aehIsolationCampaign.test.ts` 1 file / 3 tests passed; focused S10 set 5 files / 30 tests passed; typecheck exit 0; `git diff --check` clean. S10 remained pending at that checkpoint; the subsequent final independent re-acceptance is recorded below.

**Final independent re-acceptance — 2026-10-01:** the fresh S10 reviewer returned `SLICE_ACCEPTED` on reviewed head `87144f77a0c99088d385e848925643b5439422ea` ([review artifact](evidence/s10/review-s10-final.json), SHA-256 `f940aa6a72587bccf800ceade409de9b22c175af21fa83b2b99cd63e3362a9cc`). The review independently verified the repaired HOME fixture, source-impact analysis, hosted rootless execution, clean candidate-bound real Trivy evidence, stale/tampered evidence rejection, fail-closed provider behavior, unchanged TARGET, and unchanged S14/closure inputs. The exact hosted rootless job was job `110435254361` in [run 36881830507](https://github.com/JamesMorales04/agentic-engineering-harness/actions/runs/36881830507), with 3/3 real isolation tests passing and no skip/continue-on-error path. Clean candidate-bound SAST evidence is [real-sast-campaign-85877349558477e21386d6e076393d15e2a90e26-release-1790866583369-387000-cc8194fb.json](evidence/s10/real-sast-campaign-85877349558477e21386d6e076393d15e2a90e26-release-1790866583369-387000-cc8194fb.json), SHA-256 `36a871fd9c8fcafdbb4c116b1d2974fa385ab17e510ba74bdbbf6a430b8e3641`; the build identity is clean (`dirty:false`) and bound to implementation SHA `85877349558477e21386d6e076393d15e2a90e26`. The independent reviewer found no material S10 product-contract change and no reason to rerun S14 or closure.

### S11 — Contract / Integration / Browser / Visual Validation

**Objective:** produce candidate-bound Pact/OpenAPI/BDD, isolated-service integration, browser, and visual evidence, with required-tool absence failing closed and environments cleaning up.

**Dependencies:** S6 acceptance/evidence contracts, S9 Control Center/runtime, and S10 isolation. Adapters may be developed earlier; certification waits for stable interfaces.

**Gate:** required contract and UI assertions resolve to approved providers and current-candidate evidence; missing providers block; browser and visual journeys are reproducible and traceable.

**Campaign status:** **INDEPENDENTLY ACCEPTED (R3, 2026-09-26; checkpoint commit 78de9c5).** The first submission was rejected (F1–F5) and the fresh re-review (R2) verified those repairs but rejected the slice on the declared BDD/gherkin route and the LOW artifact-durability finding; both round-2 repairs were independently verified by a third fresh reviewer (R3) with all 18 S11 gate items passing, including the R2-A BDD provider route and R2-B prior-candidate artifact durability. The S11 WorkGraph, frozen lane-evidence contract, delegation assessment, exact environment facts, and pre-repair execution record plus the round-1 and round-2 repair WorkGraphs/evidence are in [S11 WorkGraph](CORE_ARCHITECTURE_V2_S11_WORKGRAPH.md). The pre-S11 gaps were independently reproduced by a bounded worker in an isolated Paseo worktree: [baseline gaps](evidence/s11/baseline-gaps.md) (unchanged). The first rejected run's machine summary is preserved as `docs/evidence/s11/browser-visual-campaign.rejected-pre-repair.json`.

**Candidate-bound provider lane evidence — 2026-09-26 (round-2 remediated):** `src/validation/laneEvidence.ts` adds `ProviderLaneEvidenceV1` for CONTRACT/INTEGRATION/BROWSER/VISUAL lanes: candidate and workspace source-digest binding, provider identity/version, command digest, status, findings, per-artifact digests (raw/report/screenshot/trace/video/log/diff/baseline), timestamps, explicit blockers, optional VISUAL comparison configuration, and a canonical digest under a candidate-scoped path. `persist`/`load`/`verify`/`require` reject workspace drift, wrong-candidate (stale), tampered JSON, tampered raw output, and missing/altered artifacts; lanes are separate and one lane's artifact cannot satisfy another. The first review found that the S4 impact path synthesized lane evidence from any passing project-script/configured-command check; the synthesis helper was deleted and `requireProviderLaneEvidenceForActionV1` now loads and verifies the evidence persisted by the matching `contract-test`/`bdd`/`integration-test`/`browser-test`/`visual-test` provider/adapter execution, failing closed with `PROVIDER_LANE_EVIDENCE_REQUIRED`. The OpenAPI adapter persists its own candidate-bound CONTRACT evidence. The R2 review then found the declared `bdd`/`gherkin` adapter still bypassed the capability provider (silent SKIP coerced to FAIL with no typed blocker, configured BDD providers ignored, no candidate-bound CONTRACT evidence); `src/validators/registry.ts` now routes `gherkin`/`bdd` through `runCapabilityValidator` and the superseded `src/validators/gherkin.ts` adapter is deleted, with the maturity inventory migrated to `src/validators/registry.ts`. Deterministic coverage: `tests/validation/laneEvidence.test.ts` 7 tests.

**Fail-closed provider resolution — 2026-09-26 (round-2 remediated):** declared `browser-test`/`visual-test`/`static-security`/`dependency-security` capabilities now resolve to matching adapters (`playwright`/`visual`/`opengrep`/`trivy`) instead of falling through to generic unit-test execution; unsupported declared capabilities return `UNSUPPORTED_VALIDATION_CAPABILITY`. A required Pact verifier, OCI runtime, pinned Playwright binary, or Playwright browser that is absent returns an explicit `CONTRACT_PROVIDER_UNAVAILABLE`, `INTEGRATION_PROVIDER_UNAVAILABLE`, `BROWSER_PROVIDER_UNAVAILABLE`, or `VISUAL_PROVIDER_UNAVAILABLE` FAIL — never a silent SKIP. A required declared `bdd` capability follows the same typed path: no runner → `CONTRACT_PROVIDER_UNAVAILABLE` (normalized `SKIP` retained only as provider detail), configured provider → executed and persisted as candidate-bound CONTRACT evidence, non-zero runner → FAIL. The OCI provider now probes the container runtime's server (`podman/docker info`) so a client without a reachable daemon is unavailable. The integration lifecycle requires explicit provision/readiness/test/cleanup steps, skips dependent steps after a provision failure, always attempts cleanup, records `cleanupRequired: provisioned`, and emits `INTEGRATION_CLEANUP_REQUIRED` only when resources were actually provisioned without a cleanup mechanism (a failed provision that created nothing yields no cleanup blocker). Deterministic coverage: `tests/validation/failClosedProviders.test.ts` 12 tests (including 3 BDD provider-resolution tests) plus the 8-test adversarial `tests/validation/requiredLaneEvidence.test.ts` dispatch suite.

**Real contract and integration campaign — 2026-09-26:** `tests/system/aehContractIntegrationCampaign.test.ts` 3 tests passed. A real OpenAPI baseline/current comparison ran through `runConfiguredValidators` and produced verified candidate-bound CONTRACT evidence (a breaking change fails); the real `tests/fixtures/bdd/node-runner.mjs` BDD runner ran through the capability provider and produced verified CONTRACT evidence; and a real loopback HTTP service lifecycle (`tests/fixtures/integration/http-service.mjs`) provisioned, proved readiness, ran a real health/pid test, cleaned up, and produced verified candidate-bound INTEGRATION evidence with the service process asserted terminated and its pid file removed. The R2 blocking finding is covered by three focused regressions: a declared `bdd` capability with no runner returns typed `CONTRACT_PROVIDER_UNAVAILABLE` instead of the bare `bdd-runner acceptance was skipped.` message; a configured BDD provider command runs the configured runner (the package.json `bdd` script is not substituted) and persists verifiable candidate-bound CONTRACT evidence; and a configured runner that exits non-zero fails. Pact remains unprovisioned (`pact`/`pact_verifier_cli` absent) and fails closed.

**Real browser and visual campaign — 2026-09-26 (round-2 remediated):** `npm run test:browser-visual` (build `release-1790457526805-146084-3114256d`, package 0.8.4, gitSha `9b74c04167e79748312160b6476cc5050050b4e3`, dirty) exited 0 with BROWSER and VISUAL PASS, both lanes verified by the campaign itself through `verifyProviderLaneEvidenceV1` after the full run. The reusable S9 `tests/browser/` harness drove the real built candidate through the supported public start path, real Control Center UI/server/controller, and real pinned Playwright 1.62.1 Chromium: pairing and session hardening, rendered operation card, rendered current-scope projection equal to durable operation/candidate digest/policy digest/`operationExecutionRevision`/controller epoch/requestId, rendered bounded choice, and footer build identity; the visual journey asserted rendered geometry, two-capture render stability, and a committed screenshot baseline. Each lane's Playwright output directory is now candidate-scoped (`.harness/evidence/playwright-s11-output/<lane>/<candidateId>-r<revision>-<identity12>`), so a later campaign run against a different candidate cannot delete this candidate's screenshots — the R2 LOW finding; the prior-candidate evidence for `CAND-S11-BROWSER-VISUAL` r1 `a20fd933b163` still verifies after this run. Candidate `CAND-S11-BROWSER-VISUAL` r1 identity digest `7ad7d70b2ef1ac17c407b344d2c2811f3d58533699730380564bf1701d28cc44`, workspace source digest `7e791da17e1951ae96bf1c8c26438de4510c01cae1703370e2e4b3b0fdafcd42`; the campaign's pre/post-summary digest checks are both stable and a fresh post-campaign recomputation equals that source digest. BROWSER evidence `.harness/evidence/browser/CAND-S11-BROWSER-VISUAL-r1-7ad7d70b2ef1/s11-browser-journey.json` digest `d98c455a87ddcff72214d4100fb0938628496e1407300ef3a783e3d800aef932` (raw + one screenshot, provider `playwright` 1.62.1); VISUAL evidence `.harness/evidence/visual/CAND-S11-BROWSER-VISUAL-r1-7ad7d70b2ef1/s11-visual-journey.json` digest `409aca8d6deb4758f55b437ae4a92c18a6943b4b9fca66d7d164099c2b9b6515` (raw + three screenshots + baseline artifact, provider `playwright-visual` 1.62.1), with the committed baseline `tests/browser/s11-control-center-browser-visual.e2e.ts-snapshots/s11-control-center-heading-chromium-linux.png` digest `21ec4e27bdb2aca29a3ed672c310e35533289f4636ddfe26f93badd8772565ee` and comparison `playwright-toHaveScreenshot` `{ animations: disabled, maxDiffPixelRatio: 0.05 }`. Sanitized machine summary: `.harness/evidence/s11/browser-visual-campaign.json`. This is real-browser product evidence with the S9 file-scripted deterministic Paseo provider boundary; it is not REAL_PROVIDER, packed-UI, or S13 certification. A narrow harness repair removes `PASEO_AGENT_ID`/`PASEO_PARENT_AGENT_ID` from fixture child environments so disposable operations cannot attach to the interactive session; no journey semantics changed.

**Deterministic/build evidence — 2026-09-26 (round-2 remediation):** `npm run typecheck` exit 0. Focused S11 set 9 files / 57 tests passed (including `tests/validation/requiredLaneEvidence.test.ts` 8 adversarial dispatch tests, `tests/validation/failClosedProviders.test.ts` 12 tests with the 3 new BDD regressions, and `tests/sdd.test.ts`). Relevant system lanes (scenario matrix, concurrency, isolation, contract/integration campaign) 4 files / 23 tests passed. The full non-system suite passed 177 files with 1 skipped and 1161 tests passed / 7 skipped; the only failing file is the pre-existing `tests/paseoSdkResolve.test.ts` environment drift reproduced at baseline in S10 (3 tests). `git diff --check` clean. A fresh independent re-review (R3) verified the round-2 repairs and accepted S11 on 2026-09-26 across all 18 gate items (checkpoint commit 78de9c5); the implementing lead did not accept its own slice.

**Objective:** complete candidate/execution-bound traces and metrics plus a versioned, reproducible advisory eval corpus and export path.

**Dependencies:** S1 execution identity. This slice may progress in parallel with S2-S11.

**Gate:** telemetry identity is verifiable, local export is reproducible, and evals remain advisory rather than policy or authority.

**Campaign status:** **INDEPENDENTLY ACCEPTED (S12-REVIEW-R2, 2026-09-26; checkpoint commit ff1f420).** The S12 WorkGraph, frozen correlation/export/corpus contracts, delegation assessment, execution record, and R1 repair record are in [S12 WorkGraph](CORE_ARCHITECTURE_V2_S12_WORKGRAPH.md). The first submission was rejected on gate B (F1 corpus identity did not bind the executable scenario harness; F2 the campaign canonical metric digest was not reproducible across executions) plus a LOW docs count finding; the R1 repairs were verified by a second fresh independent reviewer, which reproduced the corpus digest with its own implementation, falsified the scenario-binding test, ran the campaign twice in independent processes with identical canonical fields, and confirmed the advisory invariant. The implementing lead did not accept its own slice.

**Deterministic evidence — 2026-09-26 (R1 repair applied):** `src/telemetry/identity.ts` defines `TelemetryCorrelationV1`, canonical `aeh.*` keys, the domain-separated correlation digest, derive/resolve/verify helpers, and explicit `TELEMETRY_IDENTITY_MISMATCH`/`TELEMETRY_IDENTITY_UNRESOLVED` violation markers. `recordEvent` resolves the current durable operation identity at emission time and never silently re-attributes a supplied stale identity; metric recording APIs verify a caller-supplied correlation against current durable truth and record nothing on mismatch (a detached observation with no resolvable durable operation keeps the supplied identity and is documented as unverified). `src/telemetry/metrics.ts` creates real OTel counters/histograms for operation, participant, runtime-session, and validation signals, wires them at production run finish, participant lifecycle, Paseo session binding, and validation finish, and exports canonical local snapshots. The `evals/corpus/corpus.json` manifest plus five cases (deterministic validation pass/fail-closed, scope governance, context budget projection, review-convergence thresholds), fixed fixtures, and deterministic scenario harnesses ship with recorded corpus/case digests and deterministic scoring. The R1 repair binds each case's executable scenario harness closure (declared scenario file plus relative imports inside `evals/`) into the case/corpus identity and records the running build identity on every result; a missing declared scenario now fails closed with `EVAL_CORPUS_SCENARIO_MISSING`. The digest rule is canonical: the unique file set is keyed by normalized repo-relative path, sorted, and hashed as `relativePath\0fileSha256` lines joined by newline, so an independent implementation reproduces it exactly. `src/evals/advisoryInvariant.ts` plus `tests/evals/advisoryInvariant.test.ts` prove no authority decision module imports eval/telemetry modules, fabricated eval/telemetry artifacts cannot change a validation gate result or durable operation truth, forged eval acceptance claims are not loadable as an acceptance disposition, and routing is unchanged. The deterministic campaign `npx tsx scripts/s12ObservabilityCampaign.ts` uses a fixed repo-relative ignored scenario root (`.harness/s12-campaign-scenario`), so candidate/policy/correlation identity and the canonical metric snapshot digest are identical across independent executions; six independent executions in separate processes produced the same canonical metric snapshot digest `838305e6fba955d5a55d64f6252909130cf7b8f005eff8eb22388d16cd522878`, candidate identity `698e397a9d2f91bde5fa0a090a4c69062e1236bc33d1dc9c9f512d38d4128de3`, policy digest `b1c8f1dd4c74e952ded2a8537fce24ef1edefdb1b6a4f22bf828f9a20d0d2ecd`, participant correlation digest `20af5760d1f4b6c7a600a259e9cb0bdf9f3ab3bae84fe3be5d9f6ce6fed5d796`, base correlation digest `ad2d03c5d1d9427b70c21f686801de7d975fe842fb46129ab9e939ab654879f3`, and corpus digest `11e9b972994e316c853893e9744109e146ffad96563ff3924dd8632b0e5485a6`. The regenerated artifact `docs/evidence/s12/s12-observability-evals-campaign.json` (SHA-256 `285e78ff74ca9bfe64bd3ece8d4018387e4508ba2597fbad94a9871f867398af`; built release `release-1790462333370-272879-9c0e6a52`, gitSha `9cdbbfe30b48f5cbcdd214724768e981cb368714`, dirty) records the verified local OTLP/HTTP JSON payload, five identity-bound trace events with one stable trace ID and a detected stale-identity violation, reproducible metric digests, five eval case results with corpus/case/build identity, and the advisory-invariant checks. Focused S12 set passed 9 files / 28 tests; affected wiring set 13 files / 126 tests; `npm run typecheck` exit 0; `npm run build` exit 0; the full non-system Vitest ran 186 files and 1,195 tests: 1,185 passed / 3 failed / 7 pending (file statuses 185 passed / 1 failed in this run; the 7 pending are the opt-in real-provider contracts in `tests/providerContracts.test.ts`, reported as skipped by some reporter views). The only failures are the 3 pre-existing `tests/paseoSdkResolve.test.ts` mise environment-drift tests unchanged from S10/S11. Relevant deterministic system lanes (scenario matrix, concurrency, contract/integration) passed 3 files / 20 tests; the isolation campaign passed 1 file / 3 tests; the adversarial suite passed 26 tests with its two long-recorded S6 objective-completion fixture failures unchanged. This is CONTRACT/UNIT/INTEGRATION/SYSTEM_DETERMINISTIC plus a local OTLP loopback lane; it is not REAL_PROVIDER, hosted-collector, BROWSER, or PACKED_E2E certification.

### S13 — Real-Provider Certification Matrix

**Objective:** certify each claimed capability through freshly packed candidates, real provider startup/execution receipts, deterministic oracle checks, and required independent review.

**Dependencies:** each claimed capability's implementation path and relevant S6-S12 evidence.

**Gate:** each claimed row has fresh candidate/build-bound REAL_PROVIDER evidence; deterministic fixtures never substitute for provider execution; BROWSER and ADVERSARIAL lanes remain separately reported.

**Campaign status (implementing lead, 2026-09-26, round 2): `SLICE_BLOCKED` — pending fresh independent review.** The S13 WorkGraph, frozen matrix classification, delegation assessment, execution record, and blocked-row actions are in [S13 WorkGraph](CORE_ARCHITECTURE_V2_S13_WORKGRAPH.md). Round 1 repaired the packed bootstrap and added the capability-journey oracle. Round 2 fixed the real-Paseo completion path (`semanticModelDeadlineMsV1 = 300_000`; reasoning-class thinking bound LIGHT/STANDARD/DEEP -> low/medium/high with the topology as upper bound; canonical-snapshot polling in `waitForPaseoAgentHandle`; supported `direction: "tail"` timeline calls; `parseSemanticAssessmentOutputV1` typed-payload extraction; `outputJsonSchema` shipped in the assessor prompt) and migrated the default `workhorse` alias to `mimo-v2.6-flash`. All four executed real-provider lanes are ACCEPTED against freshly packed candidates in disposable `/tmp` fixtures: `startup` and `project-home`/`multi-project` with real `codex-cli 0.156.1`/`gpt-6-luna` actor sessions, and `informational` with a real packed `intent` journey through a real Paseo `opencode/opencode-go/mimo-v2.6-flash` session (`thinkingOptionId: "low"`, `status: idle`) and a durable INTENT assessment cache. The checkout tracked-file digest is identical before and after all candidate journeys (`9aecda35170ad80f…`, 668 files). Machine evidence: `docs/evidence/s13/s13-real-provider-certification.json` (SHA-256 `e30db968c9e34b98d342e9e676a6a9dc90a574280e78436953248b6bd8c38621`) with per-lane artifacts under `docs/evidence/s13/lanes/`. The 22-row matrix records 4 accepted executable rows, 17 `BLOCKED_UNAVAILABLE_PROVIDER` rows with exact required user actions (packed governed-operation campaign; real-Paseo lifecycle campaign for cancel/recovery; an authorized disposable GitHub repository; provisioning for `trivy`/`opengrep`/`podman`/`opa`/`cosign`), and 1 `NOT_APPLICABLE` row (`control-center`, BROWSER lane). The implementing lead does not accept its own slice, and no REAL_PROVIDER or PACKED_E2E claim is made for any row outside the four accepted lanes.

**Round-3 campaign status (implementing lead, 2026-09-26, round 3): `SLICE_BLOCKED` — pending fresh independent review.** Round 3 built the missing harnesses and repaired six blocking product defects. `tests/packed/s13GovernedOperationCampaign.mjs` runs the canonical `aeh operation start audit|change` path against fresh packed candidates with real Paseo participants and no deterministic stubs; the AUDIT row advanced through a real `gpt-6-luna` supervisor initialization, real reviewer sessions, configured validation and into consolidation, and the CHANGE/DIRECT row through real semantic ROUTE/STACK assessments and implementation staging, before both stopped on newly recorded defects (AEH-V2-0074 supervisor event resume residency `EXECUTION_BINDING_STALE: ContextManifest changed after session materialization`; AEH-V2-0075 transient workspace-digest mismatch plus unbounded-invalid-payload semantic output). `tests/packed/s13RealPaseoLifecycleCampaign.mjs` certifies `cancel` and `recovery` against real `opencode/opencode-go/mimo-v2.6-flash` sessions owned by real durable operations (SIGKILL owner, epoch 1→2, real inspect/stop, terminal `CANCELLED`, leases drained). `tests/packed/s13GithubEffectCampaign.mjs` created the authorized disposable PRIVATE repository `JamesMorales04/aeh-s13-round3-20260927035743-6bb5da`, froze real issue #1 and proved the `ISSUE_DRIFT` gate, and executed a real human-approved gated `git.push` (`HUMAN_REQUIRED` without a decision, `EXECUTED/SUCCEEDED` with an `ACTION_AUTHORIZATION` `APPROVE` decision, real remote branch observed); `issue import` is blocked by AEH-V2-0076 (`V2_AUTHORITY_REQUIRED` outside a managed operation), `delivery` requires an accepted packed change, and repository deletion is blocked by the missing `delete_repo` scope (AEH-V2-0077) with the exact required user action recorded. Product repairs AEH-V2-0068…0073 have focused regressions and are included in the round-3 packed candidates; the checkout tracked digest is unchanged across all campaigns and no credential is persisted. Machine evidence: round-3 governed-operations summary SHA-256 `d91cc27370f9b9a0…` (later single-lane runs rewrote `s13-governed-operations.json`; the per-lane artifacts under `docs/evidence/s13/governed-lanes/` remain the immutable blocker evidence), `docs/evidence/s13/s13-real-paseo-lifecycle.json` SHA-256 `6ec75306455c3962…` (2/2 ACCEPTED), `docs/evidence/s13/s13-github-effects.json` SHA-256 `228e3e963f48fabe…` (authority ACCEPTED, issue-driven PARTIAL, delivery BLOCKED). The implementing lead does not accept its own slice.

**Round-4 independent review (fresh reviewer, 2026-09-27): `SLICE_BLOCKED` — not accepted.** The reviewer audited the dirty candidate from source and evidence (HEAD `e3dad34`, S13 fully uncommitted; TARGET untouched; all documented evidence digests recomputed and matching), re-ran 12 focused files / 84 tests and `npm run typecheck` green, and did not inherit any prior disposition. Accepted-row sweep: `startup`, `project-home`, `multi-project`, `cancel`, `recovery` and `authority` are `UNCHANGED_EVIDENCE_VALID` (no round-4 changed module executes on their lanes; a final build-bound sweep of accepted rows is still mandatory before any checkpoint); `informational` is `STALE_REQUIRES_RERUN` (round-4 assessor changes are on its real path); `issue-driven` remains `BLOCKED`/PARTIAL (AEH-V2-0076); all other FULL_OPERATION rows remain `BLOCKED`. The review found no regression in the AEH-V2-0075/0081 untracked-identity exclusion — it is explicit, path-scoped, deterministic and untracked-only; an independent reproduction confirmed a newly created legitimate untracked file still enters candidate identity and the DIRECT ChangeSet, and only `.serena`/`graphify-out` scratch is excluded — but the mandated negative regressions are missing (R4-F1). Blocking findings: (R4-F2) the round-4 report overstates the change lane ("real implementer receipt, isolated mutation and ChangeSet assembly") against a durable artifact showing `FAILED`, 0 receipts, an unmutated candidate digest and no validation/acceptance; the same prose mis-numbers AEH-V2-0078…0081 and mis-attributes the residual change failure; a claimed watchdog trace was never preserved; (R4-F3) `finalizeOperationSession` still reads/writes operation state through the isolated execution root, so isolated DIRECT receipts cannot persist — AEH-V2-0080 is incomplete; (R4-F4) the AEH-V2-0082 audit-stall mechanism is unverified hypothesis without traces, and round 5 must reproduce and classify it before repair; (R4-F5) AEH-V2-0076 cause confirmed (`intake.ts` planner launch requires a managed operation) with a precise authority-preserving repair contract; (R4-F6/F7) completion extraction, timeout classification and the workspace-binding guard need hardening. No row is accepted; round 5 continues S13 under a fresh lead. Machine record: `docs/evidence/s13/review-round4.json`.

**Round-4 campaign status (implementing lead, 2026-09-27, round 4): `SLICE_BLOCKED` — repairs applied and exercised; budget stop.** Round 4 repaired the round-3 FULL_OPERATION blockers with focused regressions: AEH-V2-0078 supervisor event continuation binding (an explicit `continueBoundSession` turn keeps the generation's frozen binding while delivering new event context), AEH-V2-0079 isolated DIRECT sessions no longer claim the operation workspace (the implementer previously edited the bound worktree directly), AEH-V2-0080 operation artifacts resolve against the control root, AEH-V2-0081 untracked provider scratch (`.serena`, `graphify-out`) is excluded from candidate identity and ChangeSet scope, and AEH-V2-0083 assessor turns capture the completed OpenCode reply and retry once for invalid or hung non-authoritative output with a bounded repair note. The durable round-4 artifacts prove only this much: the packed change lane reached real ROUTE, STACK and CANDIDATE_IMPACT assessments and candidate r2 binding, then FAILED at implementation with 0 terminal and 0 candidate-bound participant receipts, an unmutated candidate source digest and no validation/acceptance artifact; the packed audit lane reached real supervisor initialization, reviewer sessions, validation and consolidation, persisted its audit report, but never reached a terminal operation state. The round-4 prose that claimed "a real implementer receipt, isolated mutation and ChangeSet assembly" (and the corresponding ledger rows 0079/0080) was not supported by `governed-lanes/change-direct.json` and is corrected in round 5 (R4-F2). Neither row was accepted before the budget stop: the audit silent non-terminal exit is AEH-V2-0082 (round-5 reproduction classified it as a controller terminalization-escape plus bounded-work receipt gate defect, not the round-4 watchdog hypothesis) and the change surface remains exposed to real model-output flakiness (AEH-V2-0083). `aeh issue import` remains blocked by AEH-V2-0076 (no round-4 repair). Round-4 evidence: `governed-lanes/audit.json` SHA-256 `108b3cd5095a9422…`, `governed-lanes/change-direct.json` SHA-256 `7a51ad2d158bcd32…`; the round-3 accepted lifecycle/authority evidence is unchanged. The implementing lead stopped lane iterations per the campaign budget rule and reports per-row status; no row is claimed without a fresh candidate/build-bound real-provider receipt.

**Round-5 repair/re-run campaign (implementing lead, 2026-09-27): `SLICE_BLOCKED` — the `audit` row is ACCEPTED against a fresh packed candidate; `change`/`direct-change` is a truthful terminal FAILED; the implementing lead does not accept its own slice.** Round 5 reproduced the audit stall with `S13_GOV_KEEP=1`, per-stage timestamps and `operation.watchdog.*` trace capture, and classified the owning defect: the round-4 hypothesis (watchdog wake during consolidation) was wrong; the audit report was persisted and consolidation completed, but `return terminalizeOperation(...)` without `await` bypassed the controller catch, so the successful-terminalization gate rejection (`V2_TERMINAL_GATE_REJECTED` for the registered launch/supervision identities without receipts) killed the detached controller and left the operation RUNNING. Repairs: `await` on every controller terminal branch (failures now become durable FAILED); bounded-work receipts bind to the controller-issued participant identity while the Paseo session id remains receipt provenance; persistent supervisor generations and read-only Semantic Assessor identities are exempt from the bounded-work receipt loop; authority-bound and semantic-assessment runtime sessions no longer register second work participants; `finalizeOperationSession` routes all operation-state access through the authoritative operation state root (R4-F3); `aeh issue import` now runs through a controller-owned managed operation with controller-issued Planner authority and a deterministic intake terminal gate (AEH-V2-0076); the structured-output boundary uses typed timeout classification, canonical last-assistant extraction, traced timeline failures and a verified `direction: "tail"` enum (AEH-V2-0088); the workspace guard requires positive durable proof (R4-F7); and the untracked-identity filter is intent-to-add-aware with the mandated regressions (R4-F1). The packed `audit` lane is ACCEPTED: terminal `SUCCEEDED`, persisted report, one candidate-bound bounded participant receipt, materialized supervisor generation, configured validation PASS, two real sessions (`opencode-go/mimo-v2.6-flash` reviewer, `gpt-6-luna` supervisor), oracle PASS, `checkoutUntouched: true`, packed artifact `8e67326c784c8ff7…` / build `release-1790503184568-945008-1133190d` / gitSha `e3dad34` (`docs/evidence/s13/governed-lanes/audit.json` SHA-256 `84517c3dd6445f9ad1baee8b76221fc67babd75bd7841f615f7862d1c3740a6e`; last-invocation summary SHA-256 `2f25ebd7df6bf05a671e7c018c571dc6cfd8720c0f030d0897661299c5cc948d`, stamped round 5 and labelled last-invocation-only). The packed `change-direct` lane reached a truthful terminal `FAILED`: on the final build-bound run the STACK assessor returned unparseable output that survived the bounded retry (`SEMANTIC_ASSESSMENT_INVALID` captured in `operation.error`) before any participant launched, leaving candidate r2 bound to the baseline digest with no ChangeSet/validation/acceptance artifact (an earlier round-5 run on the same source reached the DELEGATED path with 4 candidate-bound receipts and a failing configured validation command). The row remains blocked on real model-output flakiness (AEH-V2-0083-class), not on terminality (`docs/evidence/s13/governed-lanes/change-direct.json` SHA-256 `beb0f5d606fdf1ee…`). `informational` is STALE_REQUIRES_RERUN relative to the round-5 build; `cancel`/`recovery`/`authority` remain accepted by path-equivalence but their evidence still binds a pre-round-5 build and a build-bound sweep remains mandatory before any checkpoint; `issue-driven`'s managed import path is REAL-PROVIDER VERIFIED in a stable clone of the authorized disposable repository (terminal `SUCCEEDED`, bounded Planner receipt, sealed `GH-1` contract digest, frozen snapshot digest, route `FORMAL_SDD`; `docs/evidence/s13/issue-import-round5.json` SHA-256 `41cfbc5ce0328f72…`) while the full row remains blocked on change/delivery; `delivery` remains blocked. Round-5 validation: focused S13 regressions green; `npm run typecheck` exit 0; `npm run build` exit 0; `git diff --check` clean; TARGET untouched; disposable fixtures/workspaces/`/tmp` cleaned; no credential persisted. The full-suite validation also exposed one real round-5 regression (AEH-V2-0089): the first receipt-identity model closed the context execution of a resumable launch identity (`tests/changeControllerOwnedDecision.test.ts` FAILED); the final model skips the status flip for `participant:<digest>` launch identities and maps terminal receipts by participant or session identity — the regression is green again. Remaining full-suite failures (189/196 files passed; 27 tests in 6 files) are environment/pre-existing: `tests/system/` daemon-status and `run`-kind objective-gate fixtures, `tests/paseoSdkResolve` (local mise 0.9.1 shadows the fixture SDK), and parallel-only flakes (`tests/operations` descendant-cancel, `aehScenarioMatrix`) that pass or are unrelated in isolation; they are independent of the certification lanes.

**Round-5 record corrections (round 6, 2026-09-27).** The round-5 lead paragraph above is corrected where the independent review found unsupported claims: the full-suite result is `192/196 files passed, 1,307 tests: 1,275 passed / 25 failed / 7 pending` (not 189/196 files and 27 tests in 6 files), and the four failing groups are classified per the record: `aehHumanBlackBox` (19), `aehAdversarialE2E` (2) and `paseoSdkResolve` (3) are `PRE_EXISTING_REPRODUCIBLE` at `e3dad34`, and `tests/operations` descendant-cancel was a `TEST_DEFECT` (racy fixture, product gate correct) fixed in round 6. The "earlier round-5 run reached the DELEGATED path with 4 candidate-bound receipts" claim is withdrawn as not evidence-bound (R5-F7). `authority` was not "UNCHANGED (round-3)": round 5 reran it against a new packed candidate and round 6 reran it again (R5-F1). The round-5 cleanup claim is corrected: `S13_GOV_KEEP=1` left the round-5 change-lane workspaces active; round 6 archived `wks_1a8379328b475b5e`, `wks_effa47345587b823` and `wks_7fcfc35e1d1bdcd5` plus all of its own disposable workspaces and `/tmp` staging (R5-F10).

**Round-5 independent review (fresh reviewer, 2026-09-27): `SLICE_BLOCKED`.** The reviewer verified HEAD `e3dad34` with S13 fully uncommitted, `git diff --check` clean, TARGET untouched, and reproduced the full suite (196 files: 192 passed / 4 failed; 1,307 tests: 1,275 passed / 25 failed / 7 pending). `audit` is `CURRENT_ACCEPTED` against the final source (real `codex/gpt-6-luna` supervisor and `opencode-go/mimo-v2.6-flash` reviewer sessions, report/validation/oracle PASS, `checkoutUntouched`); R4-F1/F3/F7, AEH-V2-0084/0085/0089 are verified from source with green focused regressions (23 files / 197 tests; `npm run typecheck` exit 0). The change/direct-change residual is classified `AEH_PROVIDER_ADAPTER_DEFECT`: the operation-scoped assessor runs without the provider-enforced JSON schema that the installed Paseo/OpenCode stack supports for the initial prompt, the STACK path can launch up to four real model turns, and the rejected reply is not captured — Round 6 must fix the adapter, bound the total retries at one owner, persist a bounded rejected-output fingerprint, and rerun to a real terminal `ACCEPTED`. The full-suite groups were independently classified: `aehHumanBlackBox` (19), `aehAdversarialE2E` (2) and `paseoSdkResolve` (3) are `PRE_EXISTING_REPRODUCIBLE` at `e3dad34` (daemon status-detail timeout; long-recorded objective-gate fixtures; mise SDK shadowing), and `tests/operations` descendant-cancel is a `TEST_DEFECT` racy fixture (product gate correct). Two record defects block any checkpoint: the round-5 GitHub artifacts were rewritten while the docs still cite round-3 digests and call `authority` "UNCHANGED", and the managed issue-import real-provider verification predates the final `state.ts`/`agentPrompt.ts` edits (`STALE_REQUIRES_RERUN`). A pre-existing clean-build asset-resolution defect was also reproduced (`init` fails with ENOENT without stale `dist/releases` shared assets; `npm pack` ships them). All other accepted rows are `STALE_REQUIRES_RERUN`, the matrix remains open, and Round 6 continues S13; S14 must not start. Machine record: `docs/evidence/s13/review-round5.json`.

**Round-6 repair/re-run campaign (implementing lead, 2026-09-27): `SLICE_BLOCKED` — the adapter and truthfulness defects are repaired; accepted rows are re-bound to the round-6 build; `change`/`direct-change` remains a truthful terminal FAILED classified `REAL_PROVIDER_CONTRACT_LIMITATION`; the implementing lead does not accept its own slice.** Round 6 fixed the round-5 change-lane root cause (AEH-V2-0092): controller-side assessor turns now use the session-creating provider-enforced structured-output channel (assessment prompt as initial prompt plus the unchanged schema), keep no writer lease, no AEH result sink, no MCP server and no participant registration, bound the total real-model attempts at two with a single owner, and persist a bounded sanitized rejected-reply fingerprint plus the assessor session id in the failure details and `semantic.assessor.reply.rejected` traces. It also fixed two product defects the real lane then exposed — free-form `CANDIDATE_IMPACT` dimensions were fabricated into unresolvable `unit-test` requirements (AEH-V2-0090: typed canonical dimension vocabulary) and deterministic DIRECT contracts had no requirement→validator traceability (AEH-V2-0091: configured command/validator binding) — plus the pre-existing clean-build asset-resolution defect (AEH-V2-0093: `init` resolves the running release's own assets; stale shared leftovers removed) and the racy descendant-cancel fixture (AEH-V2-0094). Round-6 dispositions against the round-6 build: `audit` ACCEPTED (packed terminal `SUCCEEDED`, oracle PASS, real reviewer/supervisor sessions; `governed-lanes/audit.json` `7c893bf0ea8bea20…`); `startup`/`informational`/`project-home`/`multi-project` 4/4 ACCEPTED fresh packed candidates with real providers (`s13-real-provider-certification.json` `61532afa9695cc19…`; the `informational` lane's real assessor persisted a schema-conformant INTENT assessment through the repaired path); `cancel`/`recovery` 2/2 ACCEPTED (`s13-real-paseo-lifecycle.json` `1dbf617e2b3b1658…`); `authority` ACCEPTED (real gated push, branch `aeh-s13-authority-20260927191920`; `s13-github-effects.json` `e1c89b57bb76dfc7…`); the managed issue import reached terminal `SUCCEEDED` against the current build (`issue-import-round6.json` `de71fac2982856ad…`) while the full `issue-driven` row remains NOT ACCEPTED. `change`/`direct-change` ran four build-bound real operations and remains NOT ACCEPTED: the provider-enforced STACK assessor (`opencode-go/mimo-v2.6-flash`) returned syntactically malformed JSON in both bounded attempts (stray `}` at offset 2107; sha256 `c1d23da8…`, 5,330 bytes), recorded as `REAL_PROVIDER_CONTRACT_LIMITATION` without weakening the schema, synthesizing fields or adding retries (`governed-lanes/change-direct.json` `d05a8fc650be78f1…`). All dependent FULL_OPERATION rows remain blocked on the change trunk. Round-6 focused validation: 23 S13 files / 200 tests, `tests/buildHygiene` 6/6, `tests/changes`-family regressions green; `npm run typecheck` exit 0; `npm run build` exit 0; `git diff --check` clean; TARGET byte-untouched; every campaign `checkoutUntouched: true` (`ca509c20808e2102…`, 668 tracked files); no credential persisted. Machine record and exact digests: [S13 WorkGraph](CORE_ARCHITECTURE_V2_S13_WORKGRAPH.md) (Execution record — round 6).

 **Round-6 structured-output amendment (2026-09-27): the provisional `REAL_PROVIDER_CONTRACT_LIMITATION` was withdrawn after a bounded real-channel probe.** The probe (`docs/evidence/s13/structured-output-probe.json`, `a72260fb2b1fe9fb…`) established that Paseo passes `format: json_schema` to OpenCode, OpenCode exposes the canonical object as `info.structured` through an injected `StructuredOutput` tool (no platform input validation, `StructuredOutputError.retries = 0`), and AEH consumes the canonical object whenever it exists. The owning defects were AEH-side: the Semantic Assessor permission projection's wildcard deny suppressed the `StructuredOutput` tool (wildcard-deny fails; wildcard-deny + `StructuredOutput: "allow"` succeeds canonically), and runtime-level `structuredOutput=true` was treated as model proof while the assessor shipped on `opencode-go/mimo-v2.6-flash`, which is not certified for schema-bound structured output. Repairs: explicit `StructuredOutput` allow under the retained tool denials; a deterministic per-model certified capability registry (required level `SCHEMA_BOUND_TOOL_CALL`; certified `opencode-go/gpt-6-luna`, `deepseek-v4.1-flash`, `kimi-k3`, `glm-5.3`) with the shipped assessor default `@structured-assessor` and fail-closed resolution (no silent fallback); canonical output discipline in the assessor prompt builder as error reduction only. Regressions: `tests/semanticStructuredOutputCapability.test.ts` plus interior-stray-brace rejection. Real verification: the AEH-mediated assessor probe passed in 11.3 s on `opencode-go/gpt-6-luna` (assessment digest `5afdd632…`, real session `51014e17…`), and one bounded change/direct-change rerun reached real implementation, validation PASS, a real independent Reviewer with structured evidence, `evidence.requirement-coverage` PASS and review `ACCEPTED/CONVERGED` before terminal `FAILED` at deterministic AcceptanceOracle wiring (`VERIFICATION_VALIDATION_PATH_MISSING` for the base contract assertion and review evidence requiring binding+artifact on one participant record — AEH-V2-0099/0100); the run exceeded the 900 s lane bound and terminalized later (`governed-lanes/change-direct-late-terminal.json`, `ffb42c0e838a7a74…`). No schema, retry, or gate was weakened.

**Round-6 independent review (fresh reviewer, 2026-09-27): `SLICE_BLOCKED` — no row accepted; the structured-output repairs are confirmed and should be frozen, but the accepted rows still bind the pre-amendment build and `change`/`direct-change` remains blocked on two confirmed AEH-owned deterministic wiring defects.** The reviewer verified HEAD `e3dad34` with S13 uncommitted (55 tracked modifications / 13 untracked paths), `git diff --check` clean, `docs/CORE_ARCHITECTURE_V2.md` byte-untouched, all 17 documented round-6 evidence digests MATCH, the mandated 4 focused files / 34 tests green and `npm run typecheck` exit 0 (`docs/evidence/s13/review-round6.json`). A — the structured-output repairs are valid: the assessor projection is exactly `{"*": "deny", StructuredOutput: "allow"}` (`src/agents/permissions.ts:188-195`), the canonical level ordering and per-model certification fail closed (`src/semantic/structuredOutput.ts:16-64,84-93`, `src/semantic/assessment.ts:474`), `@structured-assessor` is `opencode-go/gpt-6-luna` while `workhorse` stays `mimo-v2.6-flash`, the discipline is prompt-level only, trailing-stray-brace recovery keeps an interior stray brace invalid, retries are single-owner (`src/participants/stack.ts:296-300,365-371`), and neither `acceptanceOracle.ts` nor `run.ts` was touched (no gate weakened). Caveat R6-F1: the 11.3 s AEH-mediated probe (`5afdd632…`/`51014e17…`) is not durably recorded (the probe file keeps the pre-fix stall and the permission/model matrices). B — the amendment rerun is durable for implementation (implementer receipt `SUCCEEDED`, candidate r2→r3 `14ce1c8b…`/`a1d2fb78…`), the real reviewer receipt/session artifact (`SUCCEEDED`/`COMPLETED`), the `review` stage `COMPLETED` and the `REJECTED` AcceptanceOracle disposition (`53436e86…`, canonically recomputed) with the exact 9 blockers; "validation PASS", "requirement-coverage PASS", "reviewer.0.reviewer PASS" and "ACCEPTED/CONVERGED" are source-gated but not directly durable (the supporting artifacts were in the removed fixture) (R6-F2, R6-F4). C — AEH-V2-0099 and AEH-V2-0100 are confirmed as AEH-owned deterministic defects with exact source chains: DIRECT compilation carries no base validation requirements (`src/core/run.ts:274`, `src/architecture/candidateAssurance.ts:206-224,353-354`), so `AC-1` resolves to no candidate-bound validation requirement (`src/architecture/acceptanceOracle.ts:133-137,329`) while the separate `requirement-coverage` matcher (`src/evidence/graph.ts:31,59-60`) still PASSes; and the execution binding (authority `participant:<digest>`, `src/operations/state.ts:1253-1258`) and the result artifact (runtime session, `src/workers/agentPrompt.ts:1238-1268`) live on different participant records, so `currentReviewEvidence` (`acceptanceOracle.ts:282,285-292,301`) can never match (session id is provenance). D — every strict gate (validation path/evidence, review strength/diversity, lead, strength, dimension coverage, certification, candidate/policy/epoch and reviewer-identity checks) must remain; the repair boundary is evidence production / contract compilation / verification resolution / participant provenance wiring, and TARGET does not contradict it (no `ARCHITECTURE_DECISION_REQUIRED`). Blocking for any checkpoint: R6-F3 — every accepted round-6 lane (audit, certification 4/4, cancel/recovery, authority, issue import) binds a build produced before the amendment source edits, and `informational`'s durable assessor identity is `@workhorse`/`mimo-v2.6-flash`, which the final source refuses as uncertified; a post-amendment build-bound sweep is required. Round 7 owns AEH-V2-0099/0100 with split-path regressions, a clean terminal change run that persists the validation/review/acceptance artifacts, and the accepted-row sweep. S14 must not start.

**Round-7 repair/re-run campaign (implementing lead, 2026-09-28): `SLICE_BLOCKED` — AEH-V2-0099/0100 are repaired and a real packed lane's AcceptanceOracle now returns `ACCEPTED`; five further deterministic defects newly reached by that lane are repaired with focused regressions; the lane still does not reach a terminal `SUCCEEDED` because objective completion fails closed on a newly discovered structural receipt/candidate-lineage gap (AEH-V2-0106). The implementing lead does not accept its own slice.** Round 7 kept the structured-output disposition frozen and worked directly in this checkout (AEH self-development; no AEH operation against this repository). Repairs: **AEH-V2-0099** — `contractValidationRequirementsV1` compiles each frozen contract requirement's declared validators into exactly one candidate-bound ValidationRequirement whose id is the deterministic check id (`command.<id>` configured commands, configured validator ids), `resolveConfigured` recognizes the `command.<id>` convention, and the executed `candidate.assurance.validation.*` checks are merged into the report on every assurance recompilation, so base assertions resolve the exact validation evidence `evidence.requirement-coverage` already saw (no oracle gate changed; unknown ids skip and collisions fail closed). **AEH-V2-0100** — `recordParticipantReceipt` persists the receipt's accepted result artifact on the receipt participant identity (the controller-issued `participant:<digest>`) while keeping launch identities non-terminal and the session id as provenance, so the oracle's unchanged single-record predicate reads one coherent chain. **AEH-V2-0101** — objective completion derives the completion status of in-flight participants from the same candidate-bound settled receipt the successful-terminal gate consumes. **AEH-V2-0102/0105** — one bounded supervisor-consolidation correction (total ≤2 turns) covers a schema-invalid response or a superseded-finding provenance mismatch while the schema and exact-set checks still fail closed. **AEH-V2-0103** — exactly one current-candidate review evidence item per reviewer/dimension/assertion across review rounds. **AEH-V2-0104** — superseded review-round checks are replaced by the current round so a remediated candidate's report status is not poisoned. Six bounded real packed `change-direct` executions were run against fresh packed candidates (900 s bound never raised; run 4 recorded the explicit timeout, cancel rejection and a 600 s late-terminal wait; final run 735 s). The final run's AcceptanceOracle disposition is `ACCEPTED` (all three assertions covered, zero blockers, persisted artifact sha256 `c325323b156c131d6442aab407b6f0042e6ed1e262558c968723096311f487e9`), validation PASS, review `ACCEPTED/CONVERGED`, real `opencode-go/mimo-v2.6-flash` implementer/reviewer and `codex/gpt-6-luna` supervisor sessions, durable `ROUTE`/`STACK`/`CANDIDATE_IMPACT` assessments, packed artifact `8ed50f013d4984e4210145d7ca1cef6cbfbc1dd80558b2244dad7c903d84603d` / packed build `release-1790556382520-15303-fd958fbb` / source `e3dad34`, `checkoutUntouched: true` (`bfee9f4bcf471394…`, 668 files). Objective completion then failed closed with one `PARTICIPANT_INCOMPLETE`: the implementer receipt binds the pre-assembly candidate r2 while the current candidate is r3 (the assembled result of that turn's ChangeSet) — **AEH-V2-0106**, newly discovered and owned by round 8 (assembly-aware receipt binding or assembly-lineage verification; no gate weakened, no evidence re-bound). Round-7 focused validation: 42 files / 417 tests green; `npm run typecheck` exit 0; `npm run build` exit 0; `git diff --check` clean; TARGET byte-untouched; all disposable workspaces and `/tmp` staging archived; no credential persisted. R6-F1/R6-F2/R6-F4/R6-F5 are addressed by the campaign's new terminal-artifact capture and the per-campaign digest correction; R6-F3 (post-amendment build-bound sweep of accepted rows) and the dependent FULL_OPERATION rows remain open; S14 must not start. Machine record and exact per-run digests: [S13 WorkGraph](CORE_ARCHITECTURE_V2_S13_WORKGRAPH.md) (Execution record — round 7).

**Round-8 repair/re-run campaign (implementing lead, 2026-09-28): `SLICE_BLOCKED` — the change/direct-change trunk reached a real terminal `SUCCEEDED` with an `ACCEPTED` AcceptanceOracle after the AEH-V2-0106 assembly-lineage repair; AEH-V2-0107 (FORMAL_SDD escalation policy rebind) and the R7-F2/F4 campaign retention/scoping defects are repaired; the dependent `delegated-change` row and the R6-F3 sweep remain open on two newly discovered defects (AEH-V2-0108 audit-consolidation hang, AEH-V2-0110 DELEGATED wave planning failure). The implementing lead does not accept its own slice.** Round 8 worked directly in this checkout (HEAD `e3dad34`, S13 uncommitted; `git diff --check` clean; `docs/CORE_ARCHITECTURE_V2.md` byte-untouched `4954b62bfb3a820f…`; structured-output routing/schemas/retries/registry/assessor discipline frozen). **AEH-V2-0106 (FIXED, real-provider exercised)** — every successful candidate transition now records a deterministic `CandidateAssemblyReceiptV1` at the ASSEMBLING boundary binding base candidate, assembled candidate, ChangeSet/patch digests, producer identity, and (when present) the settled source receipt's exact id/canonical digest; `resolveCandidateLineageReceiptV1` proves the receipt's observed candidate lies on the walked assembly ancestry, and both objective participant accounting and the successful-terminal receipt decision consume it while `evaluateTerminalGate` stays byte-identical and is still evaluated against each receipt's own binding. The packed `change-direct` lane then reached terminal **`SUCCEEDED`** (operation `CHANGE-20260928T013113Z-5de819ea`, revision 84, 513 s inside the 900 s bound, packed artifact `c2312547b41b4266…`, build `release-1790559036086-103163-19047f5f`, source `e3dad34`): AcceptanceOracle `ACCEPTED`, objective completion complete with zero blockers, every participant snapshot `COMPLETED` (the implementer's r2 receipt closes through the persisted `assembly:…:r3` lineage), validation PASS, 3 real sessions, `checkoutProof.checkoutUntouched: true`, 4 workspaces archived with `remaining: []`, and the retained acceptance artifact bytes recompute to `4263478a961036d2…`. **AEH-V2-0107 (FIXED, unit-regressed)** — the first `change-delegated` run escalated to FORMAL_SDD/ELEVATED but left the frozen policy at DELEGATED/NONE, failing the Spec Manager with `EXECUTION_POLICY_STALE`; `rebindEscalatedChangePolicy` now deterministically advances execution semantics (with a pre-escalation baseline when no digest exists) and rebinds the escalated policy before any FORMAL_SDD participant launches, idempotently and without weakening any policy guard. **AEH-V2-0109 (FIXED)** — the campaign now scopes verification/rows to the observed route, retains per-lane checkout proof, workspace inventory/archive-query evidence and recomputable acceptance-artifact bytes, and projects `candidateAssemblyReceipts`; R7-F2/F4 are closed and R7-F3 was re-run green in this environment. **Newly discovered, OPEN:** AEH-V2-0108 — the build-bound `audit` sweep run completed validating/reviewing then hung 20+ minutes in consolidation on a real supervisor continuation turn (one bounded correction fired; the watchdog only observed `supervisor-busy`), exceeding the 900 s bound plus a 600 s late wait with the operation still `RUNNING`, so it needs a bounded supervisor-turn deadline and typed fail-closed terminalization; AEH-V2-0110 — the `change-delegated` rerun failed run-internal `planning` in ~17 s with no Implementer participant and no ChangeSet, and two bounded Repairers produced zero ChangeSets, so the DELEGATED planning path requires diagnosis (deterministic vs provider quality) before the row can pass. Round-8 focused validation: 43 files / 345 tests green (including the new `assemblyLineageCompletion` 3, `specEscalationPolicyRebind` 3, `objectiveParticipantAccounting` 12 and `objectiveCompletionLifecycle` 5); `npm run typecheck` exit 0; `npm run build` exit 0 (`release-1790560487803-148242-437c07ef`); TARGET byte-untouched; all lane workspaces archived with query evidence and every `/tmp/aeh-s13-gov-*` staging directory removed; no credential persisted. Independent acceptance pending; Round 9 owns AEH-V2-0108/0110, the remaining R6-F3 sweep rows (`startup`, `informational`, `project-home`, `multi-project`, `cancel`, `recovery`, `authority`, `audit`, managed issue import), the dependent FULL_OPERATION rows and the full `issue-driven`/`delivery` contract; S14 must not start. Machine record and digests: [S13 WorkGraph](CORE_ARCHITECTURE_V2_S13_WORKGRAPH.md) (Execution record — round 8).

**S13 final acceptance and checkpoint (2026-09-30):** the fresh final independent reviewer (Round-23 close-out) returned **`SLICE_ACCEPTED`** for the complete S13 matrix on F2-final buildDigest `599ed363223ff2dd85e17296c9f34c4134bf83e363916c4c57d7e7492ccbdf45` (preserved release copy + 892-file manifest recompute MATCH; TARGET byte-identical). Machine record: `docs/evidence/s13/review-final-r23.json` SHA-256 `fe3bc3d1b507c4c4bab36f32deabdc1c98defedcb1dd12b1006b366a19bd4f8a`. All 22 capability rows are F2-bound with exact artifacts (21 current; `control-center` is `NOT_APPLICABLE`, BROWSER-only, never substituted for REAL_PROVIDER); external-effect chains use the user authorization artifact (`272c5cb7…`) through the repaired fail-closed loader with no self-minted HumanDecisions; `git diff --check` clean; no credentials in evidence. Checkpoint: commit `0bc89c836defbb2db3482e54629885a93ee8081a` (`core-v2: complete S13 real-provider certification`) pushed to `origin/core-architecture-v2`; remote HEAD verified equal; working tree clean. Non-blocking carry to S14: AEH-V2-0130 (resource lifecycle/reconciliation) with the bounded stability campaign and observations AEH-V2-0132/0133/0134; the documented pre-existing `tests/paseoSdkResolve` mise environment drift; the disclosed fixture-level brain override used in the Round-23 repair lane.

### S14 — Self-Hosting Composite Gate

**Objective:** aggregate required versioned certification evidence and prove a disposable self-modification campaign leaves the source checkout untouched.

**Dependencies:** all required prior slice gates, especially S10-S13.

**Gate:** deterministic composite gate passes; packed candidate mutations and effects remain inside the disposable fixture; reconciliation and cleanup succeed; no model declares readiness.

**S13 round 9 (historical implementing-lead report; later independently reviewed `SLICE_BLOCKED`):** AEH-V2-0108 received a 300 s maximum per persistent-supervisor continuation and a typed fail-closed timeout; audit passed in 230.784 s. Startup, informational, project-home, multi-project, cancel, recovery, and audit evidence remain independently valid. Round-9 GitHub authority and operation-bound issue-import PASS claims are tainted by AEH-V2-0113(A); they are not compliant acceptance. Round-9 direct-change failed at its 900 s bound before the cleanup guard was added. R8-F1/R8-F2 remain fixed in campaign source.

The retained Round-9 Planner artifact says `formalizationNeed: NONE` while the operation route is FORMAL_SDD; its Explorer payload was absent. A transient REQUIRED observation was discarded and is not evidence. AEH-V2-0110 remains open. The formal lane failed with `SPEC_MANAGER_RESULT_ARTIFACT_MISSING`; `CANDIDATE_WORKSPACE_MISMATCH` was agent-reported without a direct tool receipt. Round-10 removes pre-result OpenSpec placeholder writes and reports direct expected/observed workspace digests, but the live formal path has not yet been re-exercised. AEH-V2-0111 remains open. AEH-V2-0112's terminal-and-exited archive predicate passed its deterministic regression and Round-10 direct lane; its post-campaign operation read is persisted separately.

**S13 Round 11 (implementing lead; `SLICE_BLOCKED`; independent acceptance pending):** Round 11 repaired AEH-V2-0114 (P0): all campaign self-minted HumanDecision approvals are removed; the GitHub campaign consumes an externally supplied, scoped user authorization artifact (`S13_GH_AUTHORIZATION_FILE`) through a typed loader that verifies owner/PRIVATE_ONLY/effect coverage/resource namespace, binds it to the frozen allowed-effect policy and the exact ActionIntent effect digest, and fails closed with typed errors before any `gh` command, staging directory or approval. Deterministic anti-self-authorization regressions cover absence/unreadability/out-of-scope/uncovered authorization and the positive consumption path (`tests/s13GithubCampaignPolicy.test.ts`, 9). The prospective authorization artifact gained machine-readable `scope.resourceNamePrefix`/`scope.allowedEffects` (current sha256 `272c5cb7…`); issues #2/#3 were not edited or deleted and deletion was not retried. The current-build sweep is complete on packed build `0fc216b8…` (source `e3dad34`; product source unchanged): `startup`, `informational`, `project-home`, `multi-project` 4/4 PASS (`s13-real-provider-certification.json` sha256 `1ab94ea7…`) and governed `audit` PASS (lane `f31a34f8…`, summary `370b53a2…`, `checkoutUntouched: true`). A user-authorized GitHub rerun reuses issue #1: `issue-driven` PASS (import completed) and `authority` PASS with the external-authorization provenance check (summary `e8e6f5ea…`, `checkoutUntouched: true`). The Round-11 `change-delegated` diagnostic truthfully failed at FORMAL_SDD after the route escalated with durable provenance (`harness.change.triage-escalated`, explorer `4816557c…`, planner `cba6db51…`, `formalizationNeed: NONE`): the Spec Manager returned READY (`0f85afd2…`) and OpenSpec compilation rejected the persisted content for missing delta headers. New deterministic defect AEH-V2-0115 (READY content is not validated for OpenSpec canonicality) is recorded with its required repair; AEH-V2-0110/0111 remain OPEN. Every Round-11 lane envelope carries `harnessRevisions`. Remaining blockers: `delegated-change`/`formal-sdd` (AEH-V2-0110/0111/0115), `multi-worker`, repair family, `context-handoff`, `permission-delegation`, `distributed-execution`, full issue-driven end-to-end and `delivery`. `control-center` stays BROWSER-only. S14 must not start.

**S13 Round 10 (implementing lead; `SLICE_BLOCKED`; independent acceptance pending; corrected by the Round-10 independent review and Round 11):** AEH-V2-0113(A)'s repository-target gate is root-caused and fixed: the GitHub campaign requires a pre-authorized private OWNER/REPO input before external effects, has no repository creation/deletion path, and the policy regression rejects missing inputs before `gh` or staging. **Round-11 correction (AEH-V2-0114, `HISTORICAL_POLICY_VIOLATION_CONFIRMED`):** the Round-10 authority and operation-bound issue-import PASS claims are **NOT_ACCEPTED (tainted)** — the campaign fabricated its own `APPROVE` HumanDecision (`actorId: "human:s13-github"`), created issues #2/#3 after an explicit pre-effect user denial ("Do not create a repo or issue"), and pushed the authority branch under that fabricated approval. The attempt-4 remote-ref observation is genuine but its approval provenance is not user authorization. Round 11 removes all campaign-generated approvals and consumes an externally supplied scoped user authorization artifact (typed, fail-closed, bound to the frozen allowed-effect policy and the exact ActionIntent). No delete retry or credential expansion occurred; issues #2/#3 are not edited or deleted. Delivery remains unexercised/blocked for the issue-driven fixture. The Round-10 packed direct lane is terminal `SUCCEEDED`, AcceptanceOracle `ACCEPTED`, with `change` and `direct-change` rows PASS on build `0fc216b8…`; this does not self-accept S13.

Round-10 disposition/evidence map: `docs/evidence/s13/round10-review-dispositions.json` (original SHA-256 `cdf2c2f40e9f9693dae83e3bd4f0fe378f17231103f94ffbf31f0fe443a04393`; corrected in Round 11 to mark the GitHub rerun tainted — current digest in the Round-11 WorkGraph). Key digests: Round-10 GitHub attempt 4 `6cf03b9d73a9b41a7d7deb4d97b7c735ac56dd819e9f62dcc0c469550238c878` (**NOT_ACCEPTED, tainted**); lifecycle `s13-real-paseo-lifecycle-round-10-lifecycle-round10-1.json` `7e285c57463827e7a92d5358f925ccd6bbc932d90bd2330699e354e6549ce3ff`; direct summary `887f84327d78f615d4ebdc1a3e219c36a123751b4a976994fb900eac69d4b614`; delegated diagnostic `a1c632fd71415587f6f43b3fb6b23966ff0dd18bba7b3981454c7816c98c03fd`; post-campaign direct read `c80bdbf812312eb86cecf6aa04d2bb12888d3632cfedeebcbb4ee96031413a55`. Round-10 independent review: `docs/evidence/s13/review-round10.json` SHA-256 `ab31f436327641c4e8ec326ac0f4b5e13ac1c83b9f49e7f7947b60f18ccd0d44`. Remaining S13 blockers include delegated-change completion, a live FORMAL_SDD Spec Manager result, multi-worker, repair-family, context-handoff, permission-delegation, distributed-execution, full issue-driven end-to-end, and delivery. `control-center` remains BROWSER-only. AEH-V2-0113(B) cleanup is non-blocking and will be handled manually by the user. S14 must not start.

**S13 round-15 continuation (2026-09-29):** after three deterministic repairs the packed `change-delegated` lane reached a real terminal `SUCCEEDED`: **AEH-V2-0122** (the managed Lead continuation no longer requests an AEH structured-result turn for an interactive Lead session with no result channel), **AEH-V2-0123** (resolved plan-authored base validation requirements are executed into candidate-bound report checks when no matching check exists, and the Lead prompt projects the resolved `validationRequirement(kind)=status` evidence per assertion), and **AEH-V2-0124** (settled `SUCCEEDED` receipts of non-producing roles bound to an ancestor candidate of the verified assembly chain are completion evidence, and a retried launch identity's transient `FAILED` no longer shadows its settled receipt). Final lane: AcceptanceOracle `ACCEPTED` (0 blockers), Lead evidence PASS, objective completion complete, assembly receipt r2->r3, 6 real Paseo sessions, `checkoutUntouched: true`, 4 workspaces archived / 0 remaining; lane sha256 `1ea0e5f093ee9403cccf925e6e0a779e794f87557a38a4d1a422649a2561ac09`; summary sha256 `22efabf50786ab102ee704ede0efbbb582e2ad3355236cf01625321d61e8e282`. This is an implementing-lead result; independent acceptance is pending and `delegated-change` is not self-accepted.

**S13 round 18 (fresh lead, 2026-09-29; `SLICE_BLOCKED`; independent review owns acceptance):** round 18 reproduced the canonical non-system suite exactly (203 `*.test.ts` files: 201 passed / 1 skipped / 1 failed; 1,357 tests: 1,347 passed / 7 skipped / 3 failed) and classified the only failures as the documented pre-existing `tests/paseoSdkResolve.test.ts` mise/Paseo-CLI environment drift (the resolver finds the real mise store `…/npm-getpaseo-cli/0.9.1~aube~b089781319af45f6/…`; `src/paseo/sdkResolve.ts` and the test are byte-identical to HEAD; the real store is required by the running Paseo daemon so it is recorded, not provisioned away). Real-provider row work: **`certification-repair` PASS** — the failure run persisted the canonical `CertificationFailurePacket`, a real `codex/gpt-6-luna` actor turn and one bounded real repair turn repaired the ignored fixture artifact, and the deterministic oracle re-check returned PASS with the candidate tree digest stable (`r18-certrepair-4`; lane `596d8515…`, summary `066a2893…`). **`multi-worker` produced task-DAG evidence** (`workUnits=2, waves=1`) but both parallel Implementers failed `PASEO_PROVIDER_LEASE_TAKEOVER_BLOCKED` because `workflow.planning.worktreeIsolation` is declared (and defaults true) but never consumed, so same-wave units share one workspace — **new OPEN defect AEH-V2-0129**; the single-unit control run reached terminal `SUCCEEDED`/oracle `ACCEPTED`. **New OPEN defect AEH-V2-0128:** a bounded change with a concrete two-file scope takes the DIRECT branch from semantic triage while the deterministic contract floor seals DELEGATED, failing `EXECUTION_POLICY_STALE` at supervisor initialization. **`repair`/`product-repair` did not produce a repair packet** (the real implementer satisfied the seeded failing validator in one turn; exact forcing requirement recorded). The **`issue-driven` full-journey harness** (P0-A option b: disposable clone of the user-authorized private repository, real governed change, issue-bearing delivery record, external-authorization approval bound to the exact `git.push` effect digest, real push/receipts/reconciliation) is built and exercised; run `r18-journey-4` reached implementation COMPLETED and validation/review RUNNING but was non-terminal at its 900 s bound, and runs 1/2 are truthful fixture-gap failures (`impact-review-public-api` provider gap, fixed) — no accepted candidate or delivery effect was produced this round. P1-C added the explicit `delivery.delivery-record`/`delivery.provenance` checks (exercise pending the next GitHub lane run); P1-D corrected R17-F4/F5/F6/F8/F9/F10; AEH-V2-0118 is closed. `distributed-execution` is harness-built but blocked on AEH-V2-0129; `context-handoff` and `permission-delegation` remain `BLOCKED_NO_EVIDENCE` with exact dependencies. Every round-18 lane reports its own checkout proof; all lane workspaces and `/tmp/aeh-s13-*` staging are cleaned up and 24 orphaned fixture processes were terminated. S13 stays uncommitted; no checkpoint before a final independent `SLICE_ACCEPTED`; S14 must not start. Machine record and exact digests: [S13 WorkGraph](CORE_ARCHITECTURE_V2_S13_WORKGRAPH.md) (Execution record — round 18).

**S13 rounds 19–20 (fresh leads, 2026-09-29/30; `SLICE_BLOCKED`; independent review owns acceptance):** Round 19 (pre-restart record plus the Director addendum) repaired AEH-V2-0129/0128 with focused regressions and froze **F1** (`trackedDigest 1183af86…`, 668 files; buildDigest `408b178d…`), proved cleanup truthfulness (96 fixture processes terminated, 48 workspaces archived, 507 `/tmp` staging dirs removed), and completed on F1: `multi-worker` PASS (`r19-mw-1`, workUnits=2 waves=1 conflicts=0, barrier PASS, accounting 6/6/0), `repair`+`product-repair` PASS (`r19-gov-3`), `change-multifile` PASS (AEH-V2-0128 geometry), `audit` PASS, `change-delegated` PASS, `change-formal` PASS, `change-direct` PASS, certification 4/4 PASS, lifecycle cancel/recovery PASS, and GitHub `issue-driven`+`authority`+`delivery` PASS (`r19-gh-1`, existing issue #1 only). Round 20 completed the missing rows and the campaign resource-lifecycle amendment. **P0-A `issue-driven` FULL JOURNEY PASS** (`r20-journey-8`, summary `73d56b5b…`, lane `eae2e262…`): authorized issue #1 snapshot + live drift → governed change operation `CHANGE-20260929T222434Z-9eed7a49` terminal SUCCEEDED (r3, oracle ACCEPTED, route DIRECT) → `HUMAN_REQUIRED` rejection boundary → external-authorization-derived approval bound to the exact `git.push` effect digest → real push `aeh-s13-issue-journey-1-issue-journey-greet-contract-comment@5f7d655d9048` with receipts and reconciliation, issue unchanged, `checkoutUntouched: true`, workspace accounting 6/4/2. P0-B `distributed-execution` remains **truthful FAIL**: the repaired harness now starts the worker at the advanced candidate worktree and executes its binary from the fixture root, the worker claims and runs a real job, but the receipt is FAIL with **new product finding AEH-V2-0131** — distributed jobs freeze `transport: "inherit"` and the two-phase worker session protocol supports only `direct`/`podman` (`src/workers/agentPrompt.ts:1355`); the operation stays RUNNING past the bound. P0-C `context-handoff`/`permission-delegation` advanced through three harness repairs (contract-routing assurance, contract task identity) and is blocked only by a provider Serena MCP startup flake. P1-C `[delivery-record, provenance]` token mapping confirmed from `r19-gh-1` (explicit no-artifact policy determination). Final validation on F1: focused S13 set 25 files / 207 tests passed; canonical non-system suite 203 files / 1,361 tests (1,351 passed / 3 failed / 7 pending; only the pre-existing `tests/paseoSdkResolve` drift fails); typecheck/build exit 0 (buildDigest `408b178d…`); `git diff --check` clean; TARGET sha256 unchanged; 668-file tracked digest `8444d6a1…`. The user-mandated resource-lifecycle amendment is recorded in the WorkGraph; **AEH-V2-0130** (no durable operation→resource ownership; no terminal/recovery reconciliation) is opened with source evidence, a runnable bounded stability harness (`tests/packed/s13ResourceStabilityCampaign.mjs`), and the exact repair plan; two stalled `r20-dist-2`/`r20-dist-4` operations are preserved and reported, 7 terminal candidate workspaces and 4 idle supervisor generations were archived, 99 orphan fixture processes terminated and 291 staging dirs removed. S13 stays uncommitted; no checkpoint before a fresh independent `SLICE_ACCEPTED`; S14 must not start and must include resource cleanup/reconciliation in its disposable self-hosting composite.

**S13 round 21 (fresh lead, 2026-09-30; `SLICE_BLOCKED` pending the final independent review; independent review owns acceptance):** round 21 closed the three open P0 rows on the AEH-V2-0131 repair freeze. **AEH-V2-0131 fixed and REAL_PROVIDER-exercised:** distributed dispatch now resolves/freezes a concrete worker session-materialization transport at job creation (`resolveDistributedWorkerTransport`; inherited Paseo orchestration resolves to the provider idle-session boundary `direct`, recorded on the job envelope) and fails closed for `none`/unknown; the worker re-validates and never passes `inherit` to `prepareRuntimeSession`; the worker stays stateless with respect to controller authority (the draft operation-context approach was withdrawn after the product correctly fenced it with `V2_CONTROLLER_FENCED`), the controller converts the settled remote result into the candidate-bound participant receipt, `prepareAgentExecutionIdentity`/`resolveStructuredResultProvenance` validate a propagated worker launch against the controller-issued authority + frozen blueprint when no durable operation record exists, and the ephemeral direct home projects only the exact provider auth file (0600, removed with the turn). `r21-dist-14` reached terminal **SUCCEEDED** (rev 137, oracle `ACCEPTED`, lease prepared=2/released=2/completed=2, real worker session receipt, `checkoutUntouched: true`). **`context-handoff` + `permission-delegation` PASS** (`r21-ctxperm-11`): the round-20 'Serena MCP environment flake' was four harness defects — the MCP entrypoint resolved to the harness script instead of the packed entry (`AEH_ENTRY_FILE` now pinned), non-conforming participant ids closed the resumable launch identity (now `participant:<16 hex>`), the continuation turn must reuse the materialized manifests, and the evidence reads were wrong (`receipt.executionBindingDigest`; durable `continuations/*.json`). **Delivery shell terminalized (R20-F7 closed):** `r21-journey-3` PASS — governed operation SUCCEEDED → human gate `HUMAN_REQUIRED` → external-authorization-bound approval → real push `aeh-s13-issue-journey-1-issue-journey-greet-contract-comment-r21-journey-3@c4b62e45cdea` with receipts/reconciliation, and the delivery operation was terminalized through the product (`transitionOperationToTerminal`, SUCCEEDED rev 82) so the delivery record/intents/receipts are owned by a terminal governed operation; issue #1 unchanged. **AEH-V2-0130:** the R20-F1 harness defect is fixed (reconciliation gated on proven terminality; `reconciliation-terminal-only` PASS) and the bounded stability pass `r21-resource-1` produced truthful `RESOURCE_ACCUMULATION_OBSERVED` evidence (a terminal SUCCEEDED operation left 4 workspaces + 4 idle agents + 28 provider/MCP processes; the second iteration grew the residue; ceilings 0; no monotonic-growth assertion passed); round 21 additionally reclaimed 17 orphaned Paseo-managed opencode servers (~5.3 GB) with proven AEH ownership and archived their agents; the durable operation→resource ownership/terminal-recovery reconciliation architecture and S14 composite inclusion remain required. Round-21 validation: focused S13 set 27 files / 242 tests passed; canonical non-system suite 203 files / 1,363 tests (1,352 passed / 4 failed / 7 pending) where three failures are the documented pre-existing `paseoSdkResolve` environment drift and `buildHygiene` failed only under concurrent suite load and passes in isolation (6 tests); `npm run typecheck` and `npm run build` exit 0 (buildDigest `599ed363223ff2dd85e17296c9f34c4134bf83e363916c4c57d7e7492ccbdf45`, matching the `r21-dist-14` packed build); `git diff --check` clean; TARGET byte-identical. Record corrections: R20-F2 AEH-V2-0128 is FIXED_AND_REAL_PROVIDER_EXERCISED (`change-multifile` `r19-gov-3`); R20-F3 the 99-process round-20 claim is restated — 19 processes are durably inventoried (`distributed-2-orphan-reconciliation.json`) and 8 more were released by archiving idle supervisor generations (`resource-reconciliation.json`); the 51/29 passes were never persisted and are withdrawn as evidence-backed counts; R20-F5 the round-20 P0-C history is 'three attempts / two harness repairs' (context-permission-round-20-r20-ctxperm-1..3); R20-F6 the round-20 census value `archiveResults.archivedAgents` (64) is authoritative (enumerated per-agent results; the 65 was a pre-loop projection and the artifact stays byte-identical); R20-F8 `certification-repair`'s pre-F1 lane is accepted by an explicit path-equivalence determination (the F1-changed modules `routingV2`/`waveExecutor`/`triage`/`runtimeCore`/`providerLifecycle` are not imported by `src/certification/**` and the campaign drives `certify.CodexAgentProvider` directly). S13 remains uncommitted; no checkpoint until a fresh independent final `SLICE_ACCEPTED`; S14 must not start.
**S13 rounds 22–23 (fresh leads, 2026-09-30; `SLICE_BLOCKED` pending the fresh final independent review; independent review owns acceptance):** the final-review `SLICE_BLOCKED` verdict (FINAL-F1/F2 staleness on the F2 product freeze) was accepted as the sweep contract. Round 22 closed the stale matrix rows on **F2-final** (`trackedDigest 3f9b7fd8…`, 668 files; buildDigest `599ed363223ff2dd85e17296c9f34c4134bf83e363916c4c57d7e7492ccbdf45`; HEAD `e3dad34`): certification 4/4, lifecycle cancel/recovery 2/2, GitHub `issue-driven` (incl. operation-bound import) + `authority` + `delivery` 3/3 (`r22-gh-1`, FINAL-F5 closed), full `issue-driven` journey with terminalized delivery shell (`r22-journey-1`, FINAL-F6 addressed via the re-read `deliveryRecordDigest bef2fe55…`), context-handoff + permission-delegation 2/2 (`r22-ctxperm-1`, FINAL-F2 closed), certification-repair PASS, governed `audit`/`change`/`direct-change`/`delegated-change` (`r22-gov-1`), `multi-worker` + `formal-sdd` (`r22-gov-2`), and `distributed-execution` F2-current from `r21-dist-14`. FINAL-F4 is closed by the preserved byte-identical release copy at `docs/evidence/s13/round-22/frozen-release/` (892 files, manifest sha256 MATCH, recomputes to `599ed363…` with the build algorithm excluding `build-identity.json`); FINAL-F3 is corrected to the recomputed `3f9b7fd8…`. Round 23 closed the last open rows: **`repair` + `product-repair` terminal `SUCCEEDED`/oracle `ACCEPTED` on F2-final (`r23-gov-9`)**, with the forcing geometry proven end to end (first-pass validation failure → canonical RepairPacket → canonical Repairer repair in contract scope → re-validation/re-review `PASS` → oracle re-check `ACCEPTED`; 5 changed files, 0 out of scope; `checkoutUntouched: true`; lane `20621848…`, summary `6ce06bc0…`). Round-23 diagnoses recorded: the `r22-gov-4`/`r23-gov-5` FORMAL_SDD escalation was a legitimate Planner `REQUIREMENT_CONTRADICTION` removed by fixture-only request wording; the `r22-gov-3`/`r23-gov-6` review stall is a bounded-participant liveness observation (candidate `AEH-V2-0132`); `r23-gov-8` exposed superseded-attempt objective-completion accounting (candidate `AEH-V2-0133`); the `r22-gov-1` `AEH_RESULT_PROVENANCE` failure is an OPEN OBSERVATION (`AEH-V2-0134`). Round-23 kept the repair lane on a disclosed fixture-level brain-model override (`opencode`/`opencode-go`/`gpt-6-luna`) because the default `codex` ChatGPT account was hard usage-limit blocked (`codex exec` probe + Paseo daemon error evidence); the override is recorded in the lane and summary `fixtureTopology` fields. Writable-environment reproduction: focused S13 set 27 files / 242 tests passed; canonical non-system suite 203 files / 1,363 tests (1,353 passed / 3 failed / 7 skipped; only the pre-existing `tests/paseoSdkResolve` environment drift, `buildHygiene` passed); `npm run typecheck`/`npm run build` exit 0 (fresh release buildDigest `599ed363…`); `git diff --check` clean; preserved release still recomputes `599ed363…`; TARGET byte-identical. Resource amendment: `docs/evidence/s13/round-23/resource-reconciliation.json` (dead-controller evidence for the preserved non-terminal r22/r23 operations; 5 late-terminal gov-6 workspaces archived; 38 orphan processes terminated; 13 bounded workers archived) and `round23-cleanup-after.json` (boundary census); **`AEH-V2-0130` remains OPEN and is carried to S14**, whose disposable self-hosting composite must include resource cleanup/reconciliation and the bounded repeated-operation stability campaign. S13 remains uncommitted; no checkpoint until a fresh independent final `SLICE_ACCEPTED`; S14 must not start.

**S14 — Self-Hosting Composite Gate (fresh slice lead, 2026-09-30; independently ACCEPTED 2026-09-30; product closure of AEH-V2-0130/0135 recorded at the end of this document):** S14 implements the deterministic composite gate of TARGET section 16 and proves a disposable self-modification campaign leaves the source checkout untouched. The WorkGraph, frozen facts, WorkUnits, execution log and final record are in [S14 WorkGraph](CORE_ARCHITECTURE_V2_S14_WORKGRAPH.md). `src/certification/composite.ts` defines a versioned frozen policy (21 required items covering every §16 item plus the resource-lifecycle amendment item and the disposable-campaign item; 24 artifact specifications: 20 versioned S1–S13 artifacts pinned by sha256 + 4 S14 runtime artifacts), a versioned evidence manifest bound to the policy digest, deterministic fail-closed verification (missing/unsafe path/unsupported version/policy-digest mismatch/frozen-pin drift/tampered/stale/replayed evidence), machine-readable reports, and `readinessAuthority: DETERMINISTIC_COMPOSITE` with `modelAuthority: NONE` (supplied model readiness claims are recorded and never evaluated). The composite gate result on candidate buildDigest `4c05da6ee7fe249bcad4adb730f2c08bb34867910eb512171bccb40b63203284` is `ready: true` — 21/21 required items, 46 verified artifact instances, 0 failed (`docs/evidence/s14/self-hosting-composite-report.json` sha256 `75b18b6e…`; manifest `d8c4a69d…`; policy digest `458145cb…`), byte-reproducible with the recorded `generatedAt`. The disposable campaign (`s14-campaign-2`, fresh packed candidate in a `/tmp` fixture, disclosed `S13_GOV_BRAIN_*` override because codex is usage-limit blocked) reached terminal `SUCCEEDED` / AcceptanceOracle `ACCEPTED` / rows `change`+`direct-change` PASS with 3 real Paseo sessions; the checkout tracked digest is identical before/after (`3fe7e1b7…`, 2,050 files); every mutation root is disposable (fixture root + Paseo-managed worktrees) and none is inside the checkout; cleanup accounting is complete with 0 remaining workspaces; no public/external effect was performed (`docs/evidence/s14/disposable-self-modification-campaign.json` `3d224601…`, raw lane `98d62371…`). The bounded resource-stability campaign ran two real governed operations: raw terminal-owned residue 3–5 workspaces / 3–5 agents / 28–29 processes per operation (within declared ceilings), terminality-gated reconciliation returned the baseline to 0/0/0 in both iterations and never touched a non-terminal resource (`baselineReturned: true`, `preservedIntact: true`, `postReconciliationNoGrowth: true`); **`AEH-V2-0130` remains OPEN as `OPEN_PRODUCT_DEFECT_EVIDENCED`** — the product-side durable operation→resource ownership/terminal reconciliation is still missing and is carried to post-v2 (`docs/evidence/s14/resource-stability-campaign.json` `090dace9…`). Adversarial checks pass (11/11 gate-level negative checks + product-level candidate-identity suites 18/18; `docs/evidence/s14/adversarial-negative-checks.json` `d987ff6e…`). Validation: focused S14 set 4 files / 43 tests passed; canonical non-system suite 204 files (202 passed / 1 skipped / 1 failed; 1,378 tests: 1,368 passed / 7 skipped / 3 failed — only the pre-existing `tests/paseoSdkResolve` mise drift); `npm run typecheck` exit 0; `npm run build` reproduced buildDigest `4c05da6e…`; `git diff --check` clean; TARGET byte-identical. The S14 `src/**` delta is a new inert module plus one export and is not executed by any S13-certified row path, so the S13 F2-final versioned matrix remains valid. The implementing lead does not accept its own slice: AEH must not govern this development checkout until a fresh independent reviewer returns `SLICE_ACCEPTED` and the Campaign Director checkpoints.

**Final closure — AEH-V2-0130/0135 product repair and Core Architecture v2 completion (fresh final-closure slice lead, 2026-09-30):** the fresh closure slice lead owned the product repair of AEH-V2-0130/0135 (durable operation→resource ownership + terminal/recovery reconciliation) after the independent S14 acceptance, and the fresh independent closure reviewer accepted the slice (`docs/evidence/closure/review-closure.json`, verdict `SLICE_ACCEPTED`); the review’s required documentation corrections are recorded as applied in the closure WorkGraph. The closure WorkGraph, execution log, failed-run history and exact artifact digests are in [Closure WorkGraph](CORE_ARCHITECTURE_V2_CLOSURE_WORKGRAPH.md). **Product architecture:** `src/runtime/operationResources.ts` maintains a durable per-operation resource registry (`.harness/operations/<id>/resources.json`) and a deterministic terminal/recovery reconciliation with a durable receipt (`resource-reconciliation.json`); ownership proof is exact durable identity (operation record bindings, `aeh.operation` labels minted only by this product, provider lease lifecycle, managed process handles, exact session inspection, exact registered staging paths) — never PID name or filesystem glob. Reconciliation classifies `LIVE_OWNED` (non-terminal → preserved), `TERMINAL_ORPHAN` (durably terminal + proven ownership → released/archived/stopped), and `UNKNOWN_OR_UNOWNED` (preserved and reported); repeated reconciliation of released resources is an idempotent `alreadyReconciled` state and real failures stay visible. Terminalization runs reconciliation for SUCCEEDED/FAILED/CANCELLED after the terminal transition; a restart loads a terminal operation with an incomplete registry through `executeOperation` (crash between terminalization and cleanup) and a bounded deterministic sweep runs at operation start; in-operation assessor/supervisor/participant worktrees are archived with the operation, the detached liveness monitor is registered as a `managed-process` resource and terminated at terminal, and pre-operation route-triage assessor sessions are released immediately by exact launch id (they have no operation owner). Policy-driven backpressure adds generic `orchestration.operations.resources` ceilings (default 64 owned resources / 16 concurrent provider sessions per operation) with a typed `RESOURCE_CEILING_EXHAUSTED` disposition; no machine-specific constants exist in Core and model output cannot widen a ceiling. **Focused regressions A–J** (plus failed-release visibility, in-line release idempotency and inspect-resolved workspace binding): `tests/operationResourceReconciliation.test.ts` 12/12; focused closure set 43 suites / 201 tests passed (`docs/evidence/closure/focused-regressions.json` `0d670ab4…`). **Real re-verification on the closure candidate** buildDigest `d266692df02c95bf432ac6c9578e127fb16584aa2cc18d469d2c844ed3b0a2ab` (`release-1790801178235-1161839-a11bfd3f`): the bounded resource-stability campaign (`closure-resource-6`, two real governed `change-direct` iterations) terminal **SUCCEEDED** with product receipts `cleanupComplete: true` and the census back at baseline **0/0/0** in both iterations, `campaignSideCleanup: "NONE"`, `postReconciliationNoGrowth: true`, classification `aehV2_0130/0135: CLOSED_BY_PRODUCT` (`docs/evidence/closure/resource-stability-product.json` `94ed891d…`; the retained superseded failure summaries run-4/run-5 plus the run-1..3 history entries in the closure WorkGraph record the fixed gaps); the disposable self-modification campaign (`closure-disposable-1`, real packed candidate in a `/tmp` fixture) terminal SUCCEEDED / oracle ACCEPTED / rows PASS, checkout tracked digest byte-identical before/after, mutations confined to disposable roots, product reconciliation 13/13 dispositions (11 reconciled + 2 inline, 0 failed) and the archived worktree observation proven by the product receipt (`docs/evidence/closure/disposable-self-modification-campaign.json` `17e9b926…`); the affected real-Paseo **cancel/recovery** lifecycle lane reran 2/2 PASS on the closure candidate (`docs/evidence/closure/paseo-lifecycle-cancel-recovery.json` `ead63493…`; its assertions are cancel/recovery state, not cleanup receipts — the product reconciliation receipts are the resource-stability and disposable-campaign artifacts). **S14 composite regeneration:** manifest `ff1b5b11…`, report `9b9e2134…`, policy digest `1be97df3…`, `ready: true`, 21/21 required items, 47 verified artifacts, 0 failed, `readinessAuthority: DETERMINISTIC_COMPOSITE`, `modelAuthority: NONE`; the resource-lifecycle item is now backed by PRODUCT-owned reconciliation and the composite is byte-reproducible with its recorded `generatedAt`. Adversarial/fail-closed checks rerun 11/11 gate-level + 18/18 product-level with 0 failed checks (`docs/evidence/closure/adversarial-negative-checks.json` `c100197d…`). Staleness: exact source-impact analysis is recorded in `docs/evidence/closure/staleness-analysis.json` (re-executed rows CURRENT; the remaining S13 rows recorded `STALE_REQUIRES_RERUN` for a hypothetical S13 re-certification with explicit bounded-scope reasoning, because their accepted assertions are not touched by post-terminal reconciliation and the affected lifecycle/recovery surfaces were re-executed). **Validation:** canonical non-system suite 205 files (202 passed / 1 skipped / 2 failed; 1,391 tests: 1,380 passed / 7 skipped / 4 failed) with failures classified `PRE_EXISTING_REPRODUCIBLE` (`tests/paseoSdkResolve.test.ts` mise/Paseo-CLI drift, 3) and `FLAKE_CONFIRMED` (`tests/buildHygiene.test.ts` real-build bound under full-suite load; 6/6 passed in isolation); `npm run typecheck` exit 0; `npm run build` reproduced the candidate digest; `git diff --check` clean; TARGET byte-identical `4954b62b…`. **AEH-V2-0130 and AEH-V2-0135 are FIXED/CLOSED with product-owned terminal/recovery reconciliation evidence. The closure review is `SLICE_ACCEPTED` for its documented scope; S13 and S14 have separate accepted records.** **Historical acceptance-record clarification (closure snapshot 2026-09-30):** the closure reviewer’s summary included an umbrella S1–S12 statement, but at that snapshot the S10 WorkGraph still required fresh independent re-acceptance and no S10-specific accepted review artifact existed. STATUS and the closure campaign therefore treated S10 as open; the later S10-specific `SLICE_ACCEPTED` record is linked above. Do not infer S10 acceptance from the umbrella summary or S14/closure acceptance. Non-blocking post-v2 observations remain explicitly open and out of this closure: AEH-V2-0132 (bounded participant-turn liveness), AEH-V2-0133 (superseded repair-attempt identity), AEH-V2-0134 (intermittent `AEH_RESULT_PROVENANCE`), the `tests/paseoSdkResolve` mise drift, and S14-F4 test-fixture hygiene; two closure-observed provider flakes (an assessor `SEMANTIC_ASSESSMENT_UNAVAILABLE` and a supervisor `PASEO_PROVIDER_LEASE_TAKEOVER_BLOCKED` after a transient first attempt) are recorded as truthful failed-run history and classified as environment/provider flakiness, not acceptance evidence. No commit/push/stash/reset/clean was performed by the closure lead; the Campaign Director checkpoints only after the fresh independent `SLICE_ACCEPTED`.
