[EHotwagner]

Refs #5136

This is phase 2 PR 6 of the board reconciler design (docs/coordination/2026-10-31-board-reconciler-design.md, §12 item 6): the contract amendment that the contract-gated board-v2 tend rules need (S1, S3, S3b, S4, S9, S9r, O1, E1, C1, W1d, T1d). It is a governing skill/policy change. It writes nothing to the board, and it does not wire the engine's marker read (see "Not in this PR").

Lifecycle contract (work-board/references/lifecycle-contract.md, both roots)

• New section "Board tend automatic writes". It carries <!-- fsgg:board-tend-contract/v1 -->, the only occurrence of that marker in the file. It states:
  • the intended-fields line grammar, matching the implemented V2Tend.parseIntendedLine;
  • the S9, S9r, W1d and T1d fill limits, including S9r's Root-authorship check;
  • the S3, S3b and S4 Delivery Status moves, which act only on strict linkage;
  • the bounded O1 Outcome roll-up (D3).
• Dispatch rule 3. It adds C1 Delivery closure: programme decision 9, Root's docs: board reconciler design amendment for decision 9, Delivery closure rule C1 (Refs #5136) #5150 conditions, and the last merge declaring fsgg:delivery-completes/v1. Root's manual sweep applies the same conditions until C1 is enabled. The rule also adds the S1 Done backstop (D2).
• Dispatch rule 5. It adds E1, the one attended membership add. Membership is not enrollment.
• Installed acceptance. "A merged source PR cannot establish installed acceptance" now says how C1 relates to it. A Delivery whose Done when owes a release, install or readback declares fsgg:delivery-continues/v1 on its source merges, so C1 never closes it on source alone. Root closes it under rule 3 once that effect is read back.
• Parent roll-up. "Never … automatically climb/close parents" now allows only the bounded O1 roll-up. The Status-ownership sentence in organization scope points to the section.

Status lifecycle (status-lifecycle.md, both roots)

• It names the bounded automatic Status writes as the one exception to Root-decided Status.
• It adds O1's limits to the Outcome paragraph.
• Closure by C1 and the S1 backstop follow rule 3.

Routine-PR contract

• pnext-item/SKILL.md, step 3 (both roots). Every routine PR body carries exactly one of fsgg:delivery-continues/v1 and fsgg:delivery-completes/v1. A Delivery is closed by Root's acceptance, or by C1 only after a last merge that declares completion.
• Board-currency gate design. The marker table says that C1 closes only after a declared completion, and that a Delivery owing a release, install or readback declares fsgg:delivery-continues/v1. The §2b table, the automatic-writes table and decision 2 name S1 as the Done backstop and add C1 and the tend Status moves. The §2b roll-up sentence names O1.

Generated and release files

• pnext-item is a kit item, and 0.114.1 is published (nuget.org newest 0.114.1; tag coherent-set/v0.114.1 at 6eca811). The kit-published-coherence pr-arm therefore requires a higher version.
  • FsggCoherentSetVersion moves to 0.114.2. This is source preparation only; the release is [COORD-EFF-02] F# board reconciler: derive, plan and auto-apply safe board fixes; propose the rest #5136 PR 7.
  • The CLI PackageReleaseNotes carries a 0.114.2 entry covering coord: the board-v2 tend writer, allowlisted and item-scoped, with C1 close and gate replay behind the enable file (Refs #5136) #5158, coord: migration-apply re-reads only the item of each planning initialization (Refs #5136) #5160, coord: board-v2 tend proposals to #4985 and the merge gate's advisory delivery-marker check (Refs #5136) #5161, coord: ConstructorIdentity admits a later main by its unchanged constructor closure, not commit-id equality (Refs #5113) #5147, coord: enrollment rebind-reader, a designed stage that rebinds the reader identity after a coherent upgrade (Refs #5120) #5152 and this PR.
  • Release-control source: the version is the declared subject of check-coherent-set-version.py, checked against the nuget.org feed and the kit/v* tags by the pr-arm. It has no clock and no ambient environment.
• Other generated files:
  • .agents/skills is mirrored byte-identical.
  • The driver and coordination-kit manifests and repos.lock are regenerated by generate-driver-manifest --write and repos.sh relock.
  • The writer-census digests of lifecycle-contract.md (both roots) and Directory.Build.props are rehashed; no sink kinds changed.
  • skills.CHANGELOG.md has a new entry.

Local checks (printed results)

• tests/skill-quality/run.sh: "all 60 rejection cases passed"
• skill-union-assert.sh: "29/29 byte-identical"; skill-view check: "29 declared skill(s) visible in every one of 2 root(s)"; skill-view-parity.sh: "AGREE … over 29 id(s)"
• generate-driver-manifest --check: current; generate-projections --check: current; repos.sh validate: OK (10 kit items)
• check-v1-writer-census.py --structural: PASS (73 source identities); census advisory: passes
• check-kit-published-coherence.py --pr-arm: ok (0.114.2 > 0.114.1); tests/kit-published-coherence/run.sh: all 273 legs ran
• check-coherent-set-version.py: ok (0.114.2); check-engine-release-notes.py: 4990 of 35000 characters; fixtures: engine-release-notes 32 passed, coherent-set-version 13 passed; tests/release-successor-candidate/run.py: 26 tests OK
• check-paths-coherence.py: 120 workflows agree; check-prose-citations.py: ok (674 documents); check-worker-id-attractor.py: ok; check-projection.py: ok (24 coherence ids)

Review repairs (Root's REQUEST_CHANGES at feabdc20, ada7990)

• B1. When the line declares another Status, it is proposed and S9 still fills Backlog (V2Tend.fs S9 arm).

• B2. status-lifecycle's In review row and validation line name tend's mapping:

  • S3 shows a strictly linked non-draft PR's Delivery In review;
  • a PR whose required validation is unfinished stays draft, which S3b shows as In progress.

  pnext-item step 3 keeps a PR draft until its required gates pass. No workflow skips draft PRs, so CI still runs on a draft.

• N1. fsgg:delivery-completes/v1 means nothing in Done when remains after the merge. fsgg:delivery-continues/v1 is declared on every merge after which any release, install or readback remains. This is fixed in lifecycle-contract, pnext-item and the gate design's marker table.

• N3. C1 reads the most recent strictly linked merge.

• N4. O1 only proposes, and writes nothing, for an Outcome with no open child, even with an empty Status. A Backlog fill of an empty Status happens only when every open child is Blocked, matching V2Tend.evalOutcomeStatus.

• N5. Section headings are matched ignoring case.

• N6. status-lifecycle names the existing automatic writers: the merge gate's post-merge transition and "Item closed → Done".

• N7. strict-item-lifecycle §6: a strict PR carries no declaration, so C1 never closes after one.

• N8. The 0.114.2 notes name the merge gate's delivery-marker check (coord: board-v2 tend proposals to #4985 and the merge gate's advisory delivery-marker check (Refs #5136) #5161).

• N2 is not in this PR; it is board-v2 tend reads the board-tend contract marker, so gated rules can be enabled (before #5136 PR 7) #5174. The BoardV2Application.fs comment now points at board-v2 tend reads the board-tend contract marker, so gated rules can be enabled (before #5136 PR 7) #5174. This is a comment-only change with no behaviour change.

• main is merged in (206dc790, then 35113805 for process: fdev lane-capacity probe, review-finding triage before merge, push before stop (Refs #5172) #5173 and Add finalizer r5 for the pod profile #5029; the changelog and driver-manifest conflicts are resolved by keeping both entries and regenerating). Mirrors, manifests, repos.lock, the census digest and the changelog are regenerated.

Not in this PR

• The engine's read of the contract marker (board-v2 tend reads the board-tend contract marker, so gated rules can be enabled (before #5136 PR 7) #5174). It must land before PR 7. Until then every gated rule stays dropped from the enable file.

architecture-map: unaffected

🤖 Generated with Claude Code

[EHotwagner]

Root/fdev: independent review of feabdc20 is REQUEST_CHANGES. Each finding is classified below.

Fix now, on this PR:

• B1: the S9 bullet must say "and S9 still fills Backlog" for other declared values, matching the design's P-INTENT and V2Tend.fs:486-489.
• B2: qualify status-lifecycle's In review row and line 30 for tend. A strictly linked non-draft PR shows In review. A PR whose required validation is unfinished stays draft, which S3b maps to In progress.
• N1: completes means "nothing in Done when remains after this merge". Declare continues on every merge after which any release, install or readback remains.
• N3: say "most recent strictly linked merge".
• N4: clarify that O1 only proposes for an Outcome with no open child.
• N5: headings are matched ignoring case.
• N6: name the merge gate's post-merge transition and "Item closed → Done" as the existing automatic writers.
• N7: update the strict-item-lifecycle wording.
• N8: the 0.114.2 release notes mention the delivery-marker check from coord: board-v2 tend proposals to #4985 and the merge gate's advisory delivery-marker check (Refs #5136) #5161.

Follow-up: N2 (engine marker read) is filed as #5174. It must land before PR 7.

A delta re-review follows. Refs #5136

[EHotwagner]

Root/fdev: the independent delta re-review APPROVED exact head 35113805.

• B1, B2 and N1–N8 are fixed. The BoardV2Application.fs comment points at board-v2 tend reads the board-tend contract marker, so gated rules can be enabled (before #5136 PR 7) #5174.
• All 29 mirrors are byte-identical and the digests are regenerated.
• skill-quality 60/60, kit coherence ok, 0.114.2, and the Release build is clean.

Finding triage: the three wording observations (the line-80 scope, pnext's draft rule, the changelog summary) change no rule's effect. They are follow-up #5185, deferred so they don't cost another full CI cycle under the current congestion (#5182).

Auto-merge is set, pinned to this head, once required CI is green. Refs #5136

[EHotwagner]

Root/fdev: merged as 84f3bb8a at the approved head 35113805. All required contexts were green. The non-required engine, pr-arm, determinism, probe, parse and others were still queued under hosted-runner congestion (#5182). They will run on main. Refs #5136