diff --git a/docs/archive/reviews/2026-07-06-p04-p12-target-state-review.md b/docs/archive/reviews/2026-07-06-p04-p12-target-state-review.md index 0ce674f..746eb66 100644 --- a/docs/archive/reviews/2026-07-06-p04-p12-target-state-review.md +++ b/docs/archive/reviews/2026-07-06-p04-p12-target-state-review.md @@ -355,3 +355,36 @@ all four were not delivered. coordinator's reported failing number two branches below it. Both are consistent with a gate that has effectively zero headroom; neither number should be treated as the authoritative figure without a fresh CI run. + +## 7. Closure resolution log (current `main` after PR #79) + +> **Added 2026-07-06 as a separated closeout log.** The audit above stays pinned to +> `a239986`. This section records the later remediation state verified on current `main` at +> `98993ff` (`test: capture evrun full smoke evidence (#79)`), using only post-audit repo +> evidence now present on `main`. + +| ID | Status | Current-main closure evidence | +| --- | ---------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | +| F1 | fixed | PR #72 / `0dfb9d0` (`test: restore branch-coverage headroom above the gate floor`) added the missing branch-coverage headroom back into the test corpus, and current [`verification.md`](../../delivery/target-state-implementation/verification.md) now codifies both the 90.5% headroom rule and the post-merge watch that this finding said were missing. | +| F2 | fixed | PR #74 / `81d5664` (`fix: support live parked decisions`) rewired the live parked decision path in `sdk.ts`, `harness.ts`, `projection.ts`, and `resume.ts`; current tests include `P09-F2: decide routes a live parked run instead of refusing` in `packages/jig-sdk/tests/sdk.unit.test.ts`. | +| F3 | partially closed | PR #78 / `dcffa56` (`docs: record p05 real forge smoke rerun`) added a current-checkout real Forge/work-source rerun under [`docs/design/evidence/2026-07-06-p05-real-forge-smoke-rerun.md`](../../design/evidence/2026-07-06-p05-real-forge-smoke-rerun.md) and updated front-door status docs to qualify the shipped claim. The same evidence and current `README.md`/`AGENTS.md`/`docs/README.md` remain explicit that held-merge plus commit-status/comment block-surfacing real-effect proof is still open, so `MERGE-5` is not fully closed. | +| F4 | fixed | PR #79 / `98993ff` updated P11 from an internally contradictory blocked-capture posture to a reconciled combined-smoke posture in [`phases/11-evrun-full-evidence.md`](../../delivery/target-state-implementation/phases/11-evrun-full-evidence.md), with per-AC current dispositions that distinguish what is now met from what remains stronger-probe debt. The finding's contradiction is closed without overclaiming no-phone-home or idempotency. | +| F5 | fixed | PR #77 / `a78f67c` (`docs: harden target-state delivery tracking`) corrected the stale track header, phase-table rows, and P11/P12 status lines on current `main`, and tightened the delivery-foundation/status-flip checks so the daisy-chain failure mode is no longer the documented workflow. | +| F6 | fixed | PR #77 / `a78f67c` updated [`docs/design/contracts/driving.md`](../../design/contracts/driving.md) to the shipped 10-verb operator-control vocabulary (`start`, `preview`, `watch`, `inspect`, `ask-why`, `notice-ack`, `notice-snooze`, `decide`, `stop`, `export`), matching ADR 0033 and the current `JigOperatorControlPort`. | +| F7 | fixed | PR #77 / `a78f67c` removed the stale "resume placement is unsettled" posture from the track README and now states the settled split directly: resume lives on `JigRecoverySurface`, not on the operator-control port or MCP surface. | +| F8 | fixed | PR #75 / `1e23b5f` (`fix: exercise negative egress probe`) replaced the hardcoded `negativeEgressProbePassed: false` behavior with an exercised probe result path in `packages/jig-sdk/src/providers/real/confinement.ts`; current Phase 04 post-merge notes record that remediation explicitly. | +| F9 | fixed | PR #75 / `1e23b5f` also closed the latent substrate-validation hole: current `createRealExecutionHost()` now refuses probe substrate requests when no approved manifest is supplied and validates every declared request before exercising the probe. | +| F10 | fixed | PR #76 / `a66747d` (`feat: surface export attribution in observations`) added export attribution to inspect/ask-why and covered it with current CLI and SDK tests, including `P10-F10: custom-dir export remains attributable from inspect and run-level ask-why`. | +| F11 | fixed | PR #76 / `a66747d` added the missing second-process watch integration test (`P08-F11: watch observes a live run from a second process and then the finished record`) and updated the phase doc to state the shipped watch vocabulary honestly: LIVE posture is projected into snapshot signals rather than a literal `stalled` status string. | +| F12 | fixed | PR #77 / `a78f67c` added the missing process teeth to current delivery rules: explicit review-arrival gate, post-merge main-branch check watch, 90.5% coverage headroom, and a rewritten same-PR `merged (#N)` status-flip rule in [`verification.md`](../../delivery/target-state-implementation/verification.md). The historical incidents remain part of the audit record, but the process gap identified by F12 is now addressed in the governing docs. | + +### Remaining open debt after the remediation wave + +- F3 is only partially closed. Current docs and the P05 rerun record still say the success-path + real Forge `open-pr` smoke is proven, while held-merge and commit-status/comment block-surfacing + legs remain unit-proven or otherwise unevidenced on a real-effect path. +- P11 evidence is stronger than it was at `a239986`, but it is still not a claim that + no-phone-home or repeated-effect idempotency is proven. Current P11 docs, evidence index, root + `README.md`, repo `AGENTS.md`, and track `verification.md` all continue to treat those as open + evidence or owner-deferral work, so this closeout does not advance P13/P14 beyond what current + `main` already states.