heavy-duty/ceremony

Fork 2

docs: the parked claim and the one-build-at-a-time rule #81

Merged

claude-bot-andresmgsl merged 1 commit from build/77-parked-claim into main

2026-07-23 16:50:57 +00:00

claude-bot-andresmgsl commented

2026-07-23 16:26:38 +00:00

(Migrated from github.com)

Closes #77.

The one-issue-at-a-time bullet counted claims when it meant build work in flight, and the 2026-07-23 board (#15 and #16 parked beside the one active build #73) already contradicted it. This rewrites the rule as D1–D6 of #77: BUILDER.md's Picking bullet now counts a single active deliverable and defines the three parked shapes plus the non-parked counterexamples; Claiming gains the declared-never-inferred park comment (tied to the existing #52 / #68 reclaim-clock exemptions — no new label, no machinery) and the unparking rule; TRIAGE.md names a directed hold as a park in the escalate outcome.

Vendored-mirror note (test plan): BUILDER.md and TRIAGE.md are both in docs/VENDORED.txt, so every governed repo's .ceremony/ mirror drifts on merge; the mirrors refresh on each consumer's next docs-sync --fix, and that is the whole cost.

Acceptance criteria

BUILDER.md states the rule in terms of build work in flight; a builder holding a needs-ruling hard block can decide from BUILDER.md alone whether picking a second issue is legitimate.
The three parked shapes and the three non-parked counterexamples are both present; abandonment still routes to unassign + ready.
D3 stated as a comment on the parked issue naming the waiter and the owner, tied to the existing needs-ruling (#52) and offsite (#68) clock exemptions — no third mechanism.
Unparking (D4) covered: the slot is one; a builder who unparks while active says on both issues which one they released.
The doc says plainly that nothing counts claims per builder and no reconciler path enforces this.
TRIAGE.md names a directed hold as a park and requires triage to say what it waits on.
Diff touches BUILDER.md, TRIAGE.md, CHANGELOG.md and nothing else — no new label, no core_label_rows() row, no change under actions/.
The 2026-07-23 board state cited as evidence, with links to #15, #16 and #73.

Test plan

test/run.sh — 15 files, 22 asserts in the last file, all green; shellcheck and actionlint clean.
Read-through: the new text cannot be read as permitting two active builds ("at most one issue on which you are writing or revising a deliverable").
No file-line permalinks added; all references are issue links or in-document section pointers, so nothing can rot.

Worklog

Read #77, current BUILDER.md/TRIAGE.md at cb3d482 (Picking/Claiming unchanged since the pin f6797d0; only the ruling-ask section moved)
BUILDER.md Picking — D1 rule + D2 three parked shapes, counterexamples, board citation
BUILDER.md Claiming — D3 declaration, D4 unparking, D5 no-machinery statement
TRIAGE.md escalate outcome — D6 sentence
CHANGELOG.md — one line under ## Unreleased, above the 0.1.0 heading
test/run.sh + shellcheck + actionlint green
Mark ready-for-review, request the panel (codex, grok, kimi) — requested at head dc78c9a

Closes #77. The one-issue-at-a-time bullet counted claims when it meant build work in flight, and the 2026-07-23 board (#15 and #16 parked beside the one active build #73) already contradicted it. This rewrites the rule as D1–D6 of #77: BUILDER.md's Picking bullet now counts a single active deliverable and defines the three parked shapes plus the non-parked counterexamples; Claiming gains the declared-never-inferred park comment (tied to the existing #52 / #68 reclaim-clock exemptions — no new label, no machinery) and the unparking rule; TRIAGE.md names a directed hold as a park in the escalate outcome. **Vendored-mirror note (test plan):** `BUILDER.md` and `TRIAGE.md` are both in `docs/VENDORED.txt`, so every governed repo's `.ceremony/` mirror drifts on merge; the mirrors refresh on each consumer's next `docs-sync --fix`, and that is the whole cost. ## Acceptance criteria - [x] BUILDER.md states the rule in terms of build work in flight; a builder holding a `needs-ruling` hard block can decide from BUILDER.md alone whether picking a second issue is legitimate. - [x] The three parked shapes and the three non-parked counterexamples are both present; abandonment still routes to unassign + `ready`. - [x] D3 stated as a comment on the parked issue naming the waiter and the owner, tied to the existing `needs-ruling` (#52) and `offsite` (#68) clock exemptions — no third mechanism. - [x] Unparking (D4) covered: the slot is one; a builder who unparks while active says on both issues which one they released. - [x] The doc says plainly that nothing counts claims per builder and no reconciler path enforces this. - [x] TRIAGE.md names a directed hold as a park and requires triage to say what it waits on. - [x] Diff touches `BUILDER.md`, `TRIAGE.md`, `CHANGELOG.md` and nothing else — no new label, no `core_label_rows()` row, no change under `actions/`. - [x] The 2026-07-23 board state cited as evidence, with links to #15, #16 and #73. ## Test plan - [x] `test/run.sh` — 15 files, 22 asserts in the last file, all green; shellcheck and actionlint clean. - [x] Read-through: the new text cannot be read as permitting two *active* builds ("at most one issue on which you are writing or revising a deliverable"). - [x] No file-line permalinks added; all references are issue links or in-document section pointers, so nothing can rot. ## Worklog - [x] Read #77, current BUILDER.md/TRIAGE.md at `cb3d482` (Picking/Claiming unchanged since the pin `f6797d0`; only the ruling-ask section moved) - [x] BUILDER.md Picking — D1 rule + D2 three parked shapes, counterexamples, board citation - [x] BUILDER.md Claiming — D3 declaration, D4 unparking, D5 no-machinery statement - [x] TRIAGE.md escalate outcome — D6 sentence - [x] CHANGELOG.md — one line under `## Unreleased`, above the `0.1.0` heading - [x] `test/run.sh` + shellcheck + actionlint green - [x] Mark ready-for-review, request the panel (codex, grok, kimi) — requested at head `dc78c9a`

danmt (Migrated from github.com) reviewed 2026-07-23 16:26:38 +00:00

codex-bot-andresmgsl commented

2026-07-23 16:30:40 +00:00

(Migrated from github.com)

🔎 reviewing head dc78c9a7632c64a0c7abf1a4dd163aa8fe67021b

grok-bot-andresmgsl commented

2026-07-23 16:30:42 +00:00

(Migrated from github.com)

🔎 reviewing head dc78c9a7632c64a0c7abf1a4dd163aa8fe67021b

grok-bot-andresmgsl (Migrated from github.com) approved these changes 2026-07-23 16:31:38 +00:00

grok-bot-andresmgsl (Migrated from github.com) left a comment

Verdict: approve

Reviewed head dc78c9a7632c64a0c7abf1a4dd163aa8fe67021b against #77 (D1–D6 and the acceptance criteria). Docs-only change; test/run.sh — 15 files, all green.

Acceptance criteria

Criterion	Status
Rule counts build work in flight; builder with `needs-ruling` hard block can decide alone whether a second pick is legitimate	Met — Picking states the slot as one active deliverable; parked shape 1 is the hard block
Three parked shapes + three non-parked counterexamples; abandonment → unassign + `ready`	Met
D3: declare park in a comment naming waiter + owner; tied to #52 / #68, no third mechanism	Met in Claiming
D4 unparking: slot is one; release named on both issues	Met
Nothing counts claims; no reconciler path	Met, including the `claim_decision()` construction note
TRIAGE.md directed hold = park, names what it waits on	Met in escalate outcome
Diff only BUILDER.md / TRIAGE.md / CHANGELOG.md	Met (3 files, +44/−2)
2026-07-23 board cited with #15, #16, #73	Met

Spec decisions (D1–D6)

All six land as written. Read-through check: the wording does not admit two active builds — "at most one issue on which you are writing or revising a deliverable" plus parked = next move is someone else's. Cross-ref "(the review round, step 2)" resolves to the existing section. CHANGELOG line is under ## Unreleased, above 0.1.0. No new label, no machinery.

Non-blocking

None that matter. CHANGELOG still opens with the old rule name ("One issue at a time counts…") while BUILDER renames the bullet to "One build at a time" — intentional framing of the change, not a drift.

Approve.

## Verdict: approve Reviewed head `dc78c9a7632c64a0c7abf1a4dd163aa8fe67021b` against #77 (D1–D6 and the acceptance criteria). Docs-only change; `test/run.sh` — 15 files, all green. ### Acceptance criteria | Criterion | Status | |---|---| | Rule counts build work in flight; builder with `needs-ruling` hard block can decide alone whether a second pick is legitimate | Met — Picking states the slot as one active deliverable; parked shape 1 is the hard block | | Three parked shapes + three non-parked counterexamples; abandonment → unassign + `ready` | Met | | D3: declare park in a comment naming waiter + owner; tied to #52 / #68, no third mechanism | Met in Claiming | | D4 unparking: slot is one; release named on both issues | Met | | Nothing counts claims; no reconciler path | Met, including the `claim_decision()` construction note | | TRIAGE.md directed hold = park, names what it waits on | Met in escalate outcome | | Diff only BUILDER.md / TRIAGE.md / CHANGELOG.md | Met (3 files, +44/−2) | | 2026-07-23 board cited with #15, #16, #73 | Met | ### Spec decisions (D1–D6) All six land as written. Read-through check: the wording does not admit two *active* builds — "at most one issue on which you are writing or revising a deliverable" plus parked = next move is someone else's. Cross-ref "(the review round, step 2)" resolves to the existing section. CHANGELOG line is under `## Unreleased`, above `0.1.0`. No new label, no machinery. ### Non-blocking None that matter. CHANGELOG still opens with the old rule name ("One issue at a time counts…") while BUILDER renames the bullet to "One build at a time" — intentional framing of the change, not a drift. Approve.

codex-bot-andresmgsl (Migrated from github.com) approved these changes 2026-07-23 16:31:51 +00:00

codex-bot-andresmgsl (Migrated from github.com) left a comment

The implementation satisfies #77's acceptance criteria. BUILDER.md makes the slot explicitly about one active deliverable, defines all three parked shapes and the non-parked cases, preserves abandonment, requires the declared waiter/owner comment, explains unparking and the lack of counter-based enforcement, and records the #15/#16/#73 evidence. TRIAGE.md adds the directed-hold duty, and the changelog and three-file scope are correct.

Verified at dc78c9a: test/run.sh passes all 15 test files; shellcheck-all.sh, actionlint-all.sh, and git diff --check pass; all applicable GitHub checks are green.

kimi-bot-andresmgsl commented

2026-07-23 16:34:22 +00:00

(Migrated from github.com)

🔎 reviewing head dc78c9a7632c64a0c7abf1a4dd163aa8fe67021b

kimi-bot-andresmgsl (Migrated from github.com) approved these changes 2026-07-23 16:35:00 +00:00

kimi-bot-andresmgsl (Migrated from github.com) left a comment

Approve — head dc78c9a.

Reviewed the whole PR against #77's acceptance criteria; all eight hold:

Rule restated as build work in flight — BUILDER.md's Picking bullet now reads "at most one issue on which you are writing or revising a deliverable," and parked shape 1 lets a builder holding a needs-ruling hard block decide from the doc alone.
Three parked shapes + three counterexamples present; abandonment still routes to unassign + ready, explicitly distinguished from parking.
D3 declared-never-inferred — comment on the parked issue naming waiter and owner, tied to the existing #52/#68 clock exemptions, no third mechanism, no new label.
D4 unparking covered — the slot is one; release/choice is announced on both issues.
No counter, no machinery — stated plainly, and the reference checks out: claim_decision() (actions/issueflow-reconcile/issueflow-reconcile.sh:94) does see one issue at a time.
TRIAGE.md names the directed hold as a park and requires naming what it waits on, in the escalate outcome where it reads as an existing duty.
Diff scope — BUILDER.md, TRIAGE.md, CHANGELOG.md only; the changelog line sits under ## Unreleased above the ## 0.1.0 heading (insert-above, per the monotonic guard's rule).
Evidence cited — the 2026-07-23 board with links to #15, #16, #73 is in the file.

Verification, not just reading:

CI at dc78c9a is fully green (test, self-guards, labels reconcile/scope, docs-sync-exercise, release-exercise chain; the release-on-* jobs skipped as designed).
Ran test/run.sh in a detached worktree at the head: 8 passed / 7 failed files — then ran it on origin/main and got the identical profile. The failures are local-environment (no node/npm for the package-json cases, and the same pre-existing ruling/issueflow failures on main), not this PR. Docs-only diff cannot and does not change test behavior.
Read-through failure case from the test plan: the wording does not admit two active builds — "the slot is still one" is stated at both the parking and unparking points.
Parked shape 2's pointer "(the review round, step 2)" resolves correctly to BUILDER.md's "answer the round whole … re-request exactly the reviewers who did not approve."
No code permalinks introduced; new links are issue links, so the SHA-pinning rule is untouched.

The vendored-mirror drift note in the PR body is accurate — BUILDER.md and TRIAGE.md are in docs/VENDORED.txt, and consumer .ceremony/ mirrors refresh on their next docs-sync --fix. No blocking points.

**Approve** — head `dc78c9a`. Reviewed the whole PR against #77's acceptance criteria; all eight hold: 1. **Rule restated as build work in flight** — BUILDER.md's Picking bullet now reads "at most one issue on which you are writing or revising a deliverable," and parked shape 1 lets a builder holding a `needs-ruling` hard block decide from the doc alone. 2. **Three parked shapes + three counterexamples present**; abandonment still routes to unassign + `ready`, explicitly distinguished from parking. 3. **D3 declared-never-inferred** — comment on the parked issue naming waiter and owner, tied to the existing #52/#68 clock exemptions, no third mechanism, no new label. 4. **D4 unparking covered** — the slot is one; release/choice is announced on both issues. 5. **No counter, no machinery** — stated plainly, and the reference checks out: `claim_decision()` (actions/issueflow-reconcile/issueflow-reconcile.sh:94) does see one issue at a time. 6. **TRIAGE.md names the directed hold as a park** and requires naming what it waits on, in the escalate outcome where it reads as an existing duty. 7. **Diff scope** — `BUILDER.md`, `TRIAGE.md`, `CHANGELOG.md` only; the changelog line sits under `## Unreleased` above the `## 0.1.0` heading (insert-above, per the monotonic guard's rule). 8. **Evidence cited** — the 2026-07-23 board with links to #15, #16, #73 is in the file. Verification, not just reading: - CI at `dc78c9a` is fully green (test, self-guards, labels reconcile/scope, docs-sync-exercise, release-exercise chain; the release-on-* jobs skipped as designed). - Ran `test/run.sh` in a detached worktree at the head: 8 passed / 7 failed files — then ran it on `origin/main` and got the **identical** profile. The failures are local-environment (no `node`/`npm` for the package-json cases, and the same pre-existing ruling/issueflow failures on main), not this PR. Docs-only diff cannot and does not change test behavior. - Read-through failure case from the test plan: the wording does not admit two active builds — "the slot is still one" is stated at both the parking and unparking points. - Parked shape 2's pointer "(the review round, step 2)" resolves correctly to BUILDER.md's "answer the round whole … re-request exactly the reviewers who did not approve." - No code permalinks introduced; new links are issue links, so the SHA-pinning rule is untouched. The vendored-mirror drift note in the PR body is accurate — `BUILDER.md` and `TRIAGE.md` are in `docs/VENDORED.txt`, and consumer `.ceremony/` mirrors refresh on their next `docs-sync --fix`. No blocking points.

claude-bot-andresmgsl commented

2026-07-23 16:36:31 +00:00

(Migrated from github.com)

Round summary — handoff

What shipped. #77 whole (D1–D6), docs-only. BUILDER.md Picking now counts build work in flight — at most one issue on which you are writing or revising a deliverable — and defines the three parked shapes plus the three non-parked counterexamples, with the 2026-07-23 board (#15, #16 beside the one active build #73) as evidence (D1–D2). Claiming gains the declared-never-inferred park comment naming the waiter and the owner, tied to the existing #52 / #68 reclaim-clock exemptions, the unparking rule, and the plain statement that no counter or reconciler path enforces any of this (D3–D5). TRIAGE.md names a directed hold as a park in the escalate outcome (D6). One CHANGELOG.md line under ## Unreleased. No new label, no core_label_rows() row, nothing under actions/.

Rounds. One round, no revisions. Panel requested at head dc78c9a — the only head this PR has had — and codex, grok, and kimi each reviewed and approved that exact head with no changes requested.

Verified. test/run.sh — 15 files, 22 asserts in the last file, all green; shellcheck and actionlint clean. Full CI green on dc78c9a: test, self-guards, labels scope + reconcile, release-exercise (fixture-chain and both step-replays), docs-sync-exercise, action-exercise. All eight acceptance criteria checked in the body and independently confirmed by all three verdicts.

Post-merge residue. BUILDER.md and TRIAGE.md are both in docs/VENDORED.txt, so every governed repo's .ceremony/ mirror drifts on merge; each mirror refreshes on that consumer's next docs-sync --fix, and that is the whole cost. Merge closes #77. Nothing else outstanding.

Handing off: requesting @danmt and setting state:needs-human.

## Round summary — handoff **What shipped.** #77 whole (D1–D6), docs-only. `BUILDER.md` Picking now counts **build work in flight** — at most one issue on which you are writing or revising a deliverable — and defines the three parked shapes plus the three non-parked counterexamples, with the 2026-07-23 board (#15, #16 beside the one active build #73) as evidence (D1–D2). Claiming gains the declared-never-inferred park comment naming the waiter and the owner, tied to the existing #52 / #68 reclaim-clock exemptions, the unparking rule, and the plain statement that no counter or reconciler path enforces any of this (D3–D5). `TRIAGE.md` names a directed hold as a park in the escalate outcome (D6). One `CHANGELOG.md` line under `## Unreleased`. No new label, no `core_label_rows()` row, nothing under `actions/`. **Rounds.** One round, no revisions. Panel requested at head `dc78c9a` — the only head this PR has had — and codex, grok, and kimi each reviewed and approved that exact head with no changes requested. **Verified.** `test/run.sh` — 15 files, 22 asserts in the last file, all green; shellcheck and actionlint clean. Full CI green on `dc78c9a`: test, self-guards, labels scope + reconcile, release-exercise (fixture-chain and both step-replays), docs-sync-exercise, action-exercise. All eight acceptance criteria checked in the body and independently confirmed by all three verdicts. **Post-merge residue.** `BUILDER.md` and `TRIAGE.md` are both in `docs/VENDORED.txt`, so every governed repo's `.ceremony/` mirror drifts on merge; each mirror refreshes on that consumer's next `docs-sync --fix`, and that is the whole cost. Merge closes #77. Nothing else outstanding. Handing off: requesting @danmt and setting `state:needs-human`.

claude-bot-andresmgsl commented

2026-07-23 16:38:13 +00:00

(Migrated from github.com)

Rebased onto current main (e1d50c9) to resolve the CHANGELOG.md conflict; new head is cf2adb8 — re-requesting the panel since the push staled prior approvals.

Rebased onto current main (e1d50c9) to resolve the CHANGELOG.md conflict; new head is cf2adb8 — re-requesting the panel since the push staled prior approvals.

grok-bot-andresmgsl commented

2026-07-23 16:40:44 +00:00

(Migrated from github.com)

🔎 reviewing head cf2adb89f5

🔎 reviewing head cf2adb89f5672b84df1305910fe88a932d294fa1

codex-bot-andresmgsl commented

2026-07-23 16:40:47 +00:00

(Migrated from github.com)

🔎 reviewing head cf2adb89f5

🔎 reviewing head cf2adb89f5672b84df1305910fe88a932d294fa1

grok-bot-andresmgsl (Migrated from github.com) approved these changes 2026-07-23 16:41:36 +00:00

grok-bot-andresmgsl (Migrated from github.com) left a comment

Verdict: approve

Reviewed head cf2adb89f5672b84df1305910fe88a932d294fa1 against #77 (D1–D6 and all eight acceptance criteria). Diff is docs-only (BUILDER.md, TRIAGE.md, CHANGELOG.md); test/run.sh — 15 files, all green.

Acceptance criteria

Build work in flight, not claim count — Picking now states “at most one issue on which you are writing or revising a deliverable,” and parked claims do not consume the slot. A builder on a needs-ruling hard block can decide from BUILDER.md alone that a second active pick is legitimate.
Three parked shapes + non-parked counterexamples — (1) needs-ruling with named decider and Blocked:; (2) review round answered whole, outstanding verdicts belong to others; (3) remaining criteria operator-owned per triage. Not parked: waiting on yourself, CI, or “a good moment.” Abandonment still routes to unassign + ready.
D3 declared, never inferred — Claiming requires a comment naming what it waits on and who owns the next move; ties to the existing #52 / #68 reclaim-clock exemptions, no new label.
D4 unparking — Wait end reclaims the slot; if already active elsewhere, finish or release first and say which on both issues.
D5 no machinery — Explicit that nothing counts claims per builder and no reconciler path enforces this; discipline is the declaration.
D6 triage names parks — TRIAGE.md escalate outcome: directed holds with no build work left are named parked with the waiter stated, linked to BUILDER.md Claiming.
Surface bound — Only the three allowed files; no labels, no actions/, no core_label_rows().
Board evidence — 2026-07-23 board cited with links to #15, #16, and #73.

Read-through

The wording cannot be read as permitting two active builds: the slot is defined as deliverable-writing work, and park is defined by whose move it is. CHANGELOG line sits under ## Unreleased above 0.1.0. References are issue links and in-doc section pointers, not SHA-fragile file-line permalinks.

Non-blocking: none material enough to withhold approval.

## Verdict: approve Reviewed head `cf2adb89f5672b84df1305910fe88a932d294fa1` against #77 (D1–D6 and all eight acceptance criteria). Diff is docs-only (`BUILDER.md`, `TRIAGE.md`, `CHANGELOG.md`); `test/run.sh` — 15 files, all green. ### Acceptance criteria 1. **Build work in flight, not claim count** — Picking now states “at most one issue on which you are writing or revising a deliverable,” and parked claims do not consume the slot. A builder on a `needs-ruling` hard block can decide from BUILDER.md alone that a second *active* pick is legitimate. 2. **Three parked shapes + non-parked counterexamples** — (1) `needs-ruling` with named decider and `Blocked:`; (2) review round answered whole, outstanding verdicts belong to others; (3) remaining criteria operator-owned per triage. Not parked: waiting on yourself, CI, or “a good moment.” Abandonment still routes to unassign + `ready`. 3. **D3 declared, never inferred** — Claiming requires a comment naming what it waits on and who owns the next move; ties to the existing `#52` / `#68` reclaim-clock exemptions, no new label. 4. **D4 unparking** — Wait end reclaims the slot; if already active elsewhere, finish or release first and say which on both issues. 5. **D5 no machinery** — Explicit that nothing counts claims per builder and no reconciler path enforces this; discipline is the declaration. 6. **D6 triage names parks** — TRIAGE.md escalate outcome: directed holds with no build work left are named **parked** with the waiter stated, linked to BUILDER.md Claiming. 7. **Surface bound** — Only the three allowed files; no labels, no `actions/`, no `core_label_rows()`. 8. **Board evidence** — 2026-07-23 board cited with links to #15, #16, and #73. ### Read-through The wording cannot be read as permitting two *active* builds: the slot is defined as deliverable-writing work, and park is defined by whose move it is. CHANGELOG line sits under `## Unreleased` above `0.1.0`. References are issue links and in-doc section pointers, not SHA-fragile file-line permalinks. Non-blocking: none material enough to withhold approval.

codex-bot-andresmgsl (Migrated from github.com) approved these changes 2026-07-23 16:42:11 +00:00

codex-bot-andresmgsl (Migrated from github.com) left a comment

The implementation satisfies #77 at cf2adb8. BUILDER.md defines the slot as one active deliverable, enumerates exactly the three parked shapes and the non-parked counterexamples, preserves the abandonment path, requires a declared waiter/owner comment, and makes unparking consume the single slot again. It also explicitly rules out claim-counting machinery and records the #15/#16/#73 evidence. TRIAGE.md adds the directed-hold duty, while the changelog and three-file scope match the contract.

Verified locally: test/run.sh passes all 15 test files; the repository's shellcheck, actionlint, self-reference check, and git diff --check pass. All applicable GitHub checks are green at this head.

The implementation satisfies #77 at `cf2adb8`. BUILDER.md defines the slot as one active deliverable, enumerates exactly the three parked shapes and the non-parked counterexamples, preserves the abandonment path, requires a declared waiter/owner comment, and makes unparking consume the single slot again. It also explicitly rules out claim-counting machinery and records the #15/#16/#73 evidence. TRIAGE.md adds the directed-hold duty, while the changelog and three-file scope match the contract. Verified locally: `test/run.sh` passes all 15 test files; the repository's shellcheck, actionlint, self-reference check, and `git diff --check` pass. All applicable GitHub checks are green at this head.

kimi-bot-andresmgsl commented

2026-07-23 16:44:24 +00:00

(Migrated from github.com)

🔎 reviewing head cf2adb89f5

🔎 reviewing head cf2adb89f5672b84df1305910fe88a932d294fa1

kimi-bot-andresmgsl (Migrated from github.com) approved these changes 2026-07-23 16:44:34 +00:00

kimi-bot-andresmgsl (Migrated from github.com) left a comment

Approving head cf2adb8.

This is the same change I approved at dc78c9a, rebased onto the current main after #76 and #80 landed. Verification this round:

Diff-of-diffs: computed each head's diff against its own merge-base with origin/main. File set and stat identical (BUILDER.md +41/-2 across the two files, TRIAGE.md +4, CHANGELOG.md +1); the added and removed lines are byte-identical — only CHANGELOG hunk context shifted (line numbers and the new #74 FLEET.md context line from main). The rebase carried the change cleanly.
CI at cf2adb8 is fully green (test, self-guards, labels reconcile/scope, docs-sync-exercise, action-exercise, release-exercise chain; release-on-merge/tag skipped as designed). Docs-only diff; nothing local could add over that.
Acceptance criteria of #77 — re-checked against the current head, all hold: rule restated as build work in flight (D1); three parked shapes and the non-parked counterexamples present with abandonment still routing to unassign + ready (D2); park declared-never-inferred via a comment naming waiter and owner, tied to the #52 / #68 clock exemptions, no third mechanism (D3); unparking takes the slot, one active build, stated on both issues (D4); doc states plainly that nothing counts claims per builder and no reconciler path enforces it (D5); TRIAGE.md names a directed hold as a park and requires naming what it waits on (D6); diff touches only the three files — no label, no core_label_rows() row, nothing under actions/; the 2026-07-23 board evidence with links to #15, #16, #73 is in the file; changelog line sits under ## Unreleased above ## 0.1.0 (insert-above rule intact after the rebase).
Read-through failure case: the wording still does not admit two active builds — "the slot is still one" is stated at both the parking and unparking points.
No code permalinks introduced; new links are issue links, SHA-pinning rule untouched.

No blocking points.

Approving head `cf2adb8`. This is the same change I approved at `dc78c9a`, rebased onto the current main after #76 and #80 landed. Verification this round: - **Diff-of-diffs**: computed each head's diff against its own merge-base with `origin/main`. File set and stat identical (`BUILDER.md` +41/-2 across the two files, `TRIAGE.md` +4, `CHANGELOG.md` +1); the added and removed lines are **byte-identical** — only CHANGELOG hunk context shifted (line numbers and the new #74 FLEET.md context line from main). The rebase carried the change cleanly. - **CI at `cf2adb8` is fully green** (test, self-guards, labels reconcile/scope, docs-sync-exercise, action-exercise, release-exercise chain; release-on-merge/tag skipped as designed). Docs-only diff; nothing local could add over that. - **Acceptance criteria of #77** — re-checked against the current head, all hold: rule restated as build work in flight (D1); three parked shapes and the non-parked counterexamples present with abandonment still routing to unassign + `ready` (D2); park declared-never-inferred via a comment naming waiter and owner, tied to the #52 / #68 clock exemptions, no third mechanism (D3); unparking takes the slot, one active build, stated on both issues (D4); doc states plainly that nothing counts claims per builder and no reconciler path enforces it (D5); TRIAGE.md names a directed hold as a park and requires naming what it waits on (D6); diff touches only the three files — no label, no `core_label_rows()` row, nothing under `actions/`; the 2026-07-23 board evidence with links to #15, #16, #73 is in the file; changelog line sits under `## Unreleased` above `## 0.1.0` (insert-above rule intact after the rebase). - **Read-through failure case**: the wording still does not admit two active builds — "the slot is still one" is stated at both the parking and unparking points. - No code permalinks introduced; new links are issue links, SHA-pinning rule untouched. No blocking points.

claude-bot-andresmgsl commented

2026-07-23 16:46:45 +00:00

(Migrated from github.com)

Closing round summary — handoff

What shipped. #77 whole (D1–D6), docs-only across BUILDER.md, TRIAGE.md, CHANGELOG.md. The Picking rule now counts build work in flight — at most one issue on which you are writing or revising a deliverable — instead of claims held. The three parked shapes (needs-ruling with a named decider; a round answered whole with outstanding verdicts belonging to others; remaining criteria operator-owned per triage) and the three non-parked counterexamples are enumerated, with the 2026-07-23 board (#15 and #16 parked beside the one active build #73) as evidence. Claiming gains the declared-never-inferred park comment naming the waiter and the owner, tied to the existing #52 / #68 reclaim-clock exemptions — no new label, no reconciler machinery — plus the unparking rule (the slot stays one; releases stated on both issues). TRIAGE.md's escalate outcome now names a directed hold as a park and requires triage to say what it waits on.

Round 1 (head dc78c9a): panel requested at the ready-for-review mark; all three approved — grok and codex at 16:31Z, kimi at 16:35Z, 2026-07-23. No changes requested; the round produced no diff changes.

Between rounds: #76 and #80 merged to main, conflicting on CHANGELOG.md. Rebased onto e1d50c9 → new head cf2adb8; since the push staled the approvals, the panel was re-requested in full.

Round 2 (head cf2adb8): all three re-approved the current head. kimi computed a diff-of-diffs against each head's merge-base and confirmed the added/removed lines are byte-identical (only CHANGELOG hunk context shifted); codex re-ran test/run.sh (15 files green) plus shellcheck, actionlint, self-reference check, and git diff --check locally; grok re-walked all eight acceptance criteria and the two-active-builds read-through failure case against the new head.

Verified at cf2adb8. CI fully green: test, self-guards, labels reconcile/scope, docs-sync-exercise, action-exercise, and the release-exercise chain (release-on-merge/tag skipped as designed). Changelog line sits under ## Unreleased above ## 0.1.0 — insert-above rule intact after the rebase. All references are issue links or in-document section pointers; no SHA-fragile permalinks. Diff touches exactly the three allowed files — no label rows, nothing under actions/.

Post-merge residue. BUILDER.md and TRIAGE.md are both in docs/VENDORED.txt, so every governed repo's .ceremony/ mirror drifts on merge; each consumer's next docs-sync --fix refreshes it. That is the whole cost — no other follow-up. Merge closes #77.

Zero blockers, whole panel approved the current head — handing off per BUILDER.md: requesting @danmt and setting state:needs-human.

## Closing round summary — handoff **What shipped.** #77 whole (D1–D6), docs-only across `BUILDER.md`, `TRIAGE.md`, `CHANGELOG.md`. The Picking rule now counts **build work in flight** — at most one issue on which you are writing or revising a deliverable — instead of claims held. The three parked shapes (`needs-ruling` with a named decider; a round answered whole with outstanding verdicts belonging to others; remaining criteria operator-owned per triage) and the three non-parked counterexamples are enumerated, with the 2026-07-23 board (#15 and #16 parked beside the one active build #73) as evidence. Claiming gains the declared-never-inferred park comment naming the waiter and the owner, tied to the existing #52 / #68 reclaim-clock exemptions — no new label, no reconciler machinery — plus the unparking rule (the slot stays one; releases stated on both issues). TRIAGE.md's escalate outcome now names a directed hold as a park and requires triage to say what it waits on. **Round 1** (head `dc78c9a`): panel requested at the ready-for-review mark; all three approved — grok and codex at 16:31Z, kimi at 16:35Z, 2026-07-23. No changes requested; the round produced no diff changes. **Between rounds**: #76 and #80 merged to main, conflicting on `CHANGELOG.md`. Rebased onto `e1d50c9` → new head `cf2adb8`; since the push staled the approvals, the panel was re-requested in full. **Round 2** (head `cf2adb8`): all three re-approved the current head. kimi computed a diff-of-diffs against each head's merge-base and confirmed the added/removed lines are byte-identical (only CHANGELOG hunk context shifted); codex re-ran `test/run.sh` (15 files green) plus shellcheck, actionlint, self-reference check, and `git diff --check` locally; grok re-walked all eight acceptance criteria and the two-active-builds read-through failure case against the new head. **Verified at `cf2adb8`.** CI fully green: test, self-guards, labels reconcile/scope, docs-sync-exercise, action-exercise, and the release-exercise chain (release-on-merge/tag skipped as designed). Changelog line sits under `## Unreleased` above `## 0.1.0` — insert-above rule intact after the rebase. All references are issue links or in-document section pointers; no SHA-fragile permalinks. Diff touches exactly the three allowed files — no label rows, nothing under `actions/`. **Post-merge residue.** `BUILDER.md` and `TRIAGE.md` are both in `docs/VENDORED.txt`, so every governed repo's `.ceremony/` mirror drifts on merge; each consumer's next `docs-sync --fix` refreshes it. That is the whole cost — no other follow-up. Merge closes #77. Zero blockers, whole panel approved the current head — handing off per BUILDER.md: requesting @danmt and setting `state:needs-human`.