diff --git a/docs/internal/cycle-ledger.md b/docs/internal/cycle-ledger.md index bb643c5144..409b06e1ec 100644 --- a/docs/internal/cycle-ledger.md +++ b/docs/internal/cycle-ledger.md @@ -66,3 +66,4 @@ earlier cycle files are records and are not rewritten. | 18 | kitchen-sink rebuild / Workforce reference epic wrap (FIX-1455) (2026-09-24) | [`cycle-18.md`](cycle-ledger/cycle-18.md) | | 19 | kitchen-sink support desk epic wrap (FIX-1592) (2026-09-29) | [`cycle-19.md`](cycle-ledger/cycle-19.md) | | 20 | hard-gates-before-public-release epic wrap (FIX-1635) (2026-10-01) | [`cycle-20.md`](cycle-ledger/cycle-20.md) | +| 21 | Shift Manager lab shell epic wrap (FIX-1649) (2026-10-04) | [`cycle-21.md`](cycle-ledger/cycle-21.md) | diff --git a/docs/internal/cycle-ledger/cycle-21.md b/docs/internal/cycle-ledger/cycle-21.md new file mode 100644 index 0000000000..5157e0b852 --- /dev/null +++ b/docs/internal/cycle-ledger/cycle-21.md @@ -0,0 +1,318 @@ +# Cycle 21 — Shift Manager lab shell epic wrap (FIX-1649) (2026-10-04) + +Part of the [cycle ledger](../cycle-ledger.md), whose header defines the feedback classes and reading labels. + +**Review was cheap. The closure did the expensive work, and it did it on the path the rules give +it.** No direction artifact took a second round, and no implementation PR took more than two fix +waves (29 waves over 29 PRs; cycle 20: 57 over 15). The rework the epic is remembered for came from +closure runs, not from review: +- three controls that went WRONG on the closure's first run (FIX-1732, FIX-1733, FIX-1734); +- a re-run on 10-04 that found three child checks and one closure leg broken by merges to `main`; +- Codex's first read of the closure runner, which found that `GOAL_PART=bogus` printed PASS with 0 + verdicts. + +Each one was found by the instrument that is meant to find it. Each one cost one fix wave or one +re-run. The three broken controls repeat cycle 19's three shapes one for one. That is the +finding this entry records. It is not yet a proposal, because the earliest point that can see each +one is the closure's own control run (see *Candidates considered and dropped*). **Proposal skipped.** + +**Method — scope.** Forty-seven reviewed artifacts. They were found from FIX-1649's 35 Linear +children (read over GraphQL) and from a listing of every repository PR opened since 09-25, matched +on the epic id and each child id in the title or branch: +- the epic PR #2421 and three amendments: #2423, #2619, and #2648 (FIX-1737's spec, which also + amends the epic); +- fourteen issue specs and spec amendments: #2424, #2425, #2426, #2428, #2434, #2436, #2437, #2438, + #2439, #2440, #2525, #2592 (the closure plan's amendment), #2614 and #2616; +- twenty-nine implementation PRs, the closure #2709 among them. + +**Out of the sample:** +- POCs #2442, #2446, #2452, #2534, #2542, #2545, #2642 and #2711. Each one is a Code Snob or + simplify subtraction that its parent PR adopted or declined. None was reviewed on its own. #2642 + is still open. +- #2420, the project spec. +- #2498 (FIX-1682), #2549 and #2569 (FIX-1701), #2603 and #2611 (FIX-1717), #2613, #2652 and #2678 + (FIX-1719), #2625 and #2687 (FIX-1718), #2633, #2643, #2661, #2686, #2701 and #2702. These name a + child in their body but belong to other issues. #2678 carries two FIX-1735 commits (`1f2dadd45` + and `f23ae7c78`) after #2654 merged. They are recorded under *Filed*, not scored. +- FIX-1671, FIX-1673, FIX-1675 and FIX-1705, children with no PR (Backlog or Todo at wrap). + +Every review, review thread, conversation comment, commit list and changed-file list was read +through the GitHub REST API. **Endpoints and rounds** follow cycle 20 unchanged: +- Direction artifacts end at the human direction approval, which is the owner's merge on all + eighteen. +- Implementation rounds are spent fix waves: pushes that answer a review, counted once per batch. +- A second-look or Architect body is read as one review body. Its items are counted only where the + author's reply names them as folded (`SL1`…), with the class of each folded item. +- A suggestion the author declined with a reason is a `nit`. A suggestion taken is classed by what + it changed. + +**Every act sits behind the shared `jhoffner` login**, and this entry does not guess who pressed +what. + +| PR | Kind | Rounds | Endpoint | Feedback classes (deduped) | Felt off? | Upstream fix that would have prevented it | +|---|---|---|---|---|---|---| +| [#2421](https://github.com/fixpoint-labs/flow-state-dev/pull/2421) epic-spec FIX-1649 | epic | ~1 (+ the owner's D3 answer and design hand-back v1, not rounds) | approval = merge 09-30 01:33Z | design-off ×1 (SL1: ER-6 and D2 collided with how `@flow-state-dev/ui` ships, so D2 now names two shelves) · spec-ambiguity ×3 (SL2: "opens by configuration alone" while trees carry code · T1+T7: the plan assumed the open fork's answer · T4: leg c open-ended) · over-engineered ×1 (SL3: a second full theme whose only consumer was the proof) · nit ×4 — *5 from 9* | no | — | +| [#2423](https://github.com/fixpoint-labs/flow-state-dev/pull/2423) epic amendment (workstream ownership) | epic (amendment) | ~1 | approval = merge 09-30 16:53Z | stale-restatement ×1 (Codex P2: the boundary summary kept the old owner) · nit ×1 — *1 from 2* | no | — | +| [#2619](https://github.com/fixpoint-labs/flow-state-dev/pull/2619) epic amendment (design v2) | epic (amendment) | ~1 | approval = merge 10-01 19:31Z, 33 min after it opened | stale-restatement ×4 (the `labs/app-lab` path after the rename · EVOLUTION still said *blocking* · an open item already answered · the path figure) · nit ×2 — *4 from 6*. **No Codex: usage limit** | no | — | +| [#2648](https://github.com/fixpoint-labs/flow-state-dev/pull/2648) spec FIX-1737 (+ epic amendment) | spec | ~1 | approval = merge 10-02 20:38Z | missed-edge-case ×4 (Codex P2 ×4: require every expected row to render (*vacuous-assertion*) · keep the CoS look check provider-free · gate completion on all three slices · run both leg-d controls in the closure) · over-engineered ×1 (two hand-kept classifiers, now `scope.json`) · nit ×5 — *5 from 10* | no | — | +| [#2424](https://github.com/fixpoint-labs/flow-state-dev/pull/2424) spec FIX-1662 | spec | ~1 | approval = merge 09-30 16:55Z | missed-edge-case ×3 (Codex P1 ×2: an identity probe · an org-wide read for every seat's asks; P2: board grouping) · over-engineered ×1 (T4+SL: PR sizing) · stale-restatement ×1 (four PRs against a three-PR figure) · nit ×3 — *5 from 8* | no | — | +| [#2425](https://github.com/fixpoint-labs/flow-state-dev/pull/2425) spec FIX-1655 | spec | ~1 | approval = merge 09-30 16:52Z | missed-edge-case ×3 (Codex P1: register every component before token dependencies · P2 ×2: literal colours in the census, the shadow-free theme) · spec-ambiguity ×1 (the denied-state token) · nit ×9 (8 Cursor notes unanswered) — *4 from 13* | no | — | +| [#2426](https://github.com/fixpoint-labs/flow-state-dev/pull/2426) spec FIX-1663 (closure plan) | spec | ~1 | approval = merge 09-30 16:44Z | missed-edge-case ×2 (**Codex P1 T9: `optimistic-post` could not isolate a4; folded as a two-leg control**, see *The class* · SL: b0's writer isolated) · nit ×8 — *2 from 10* | no | — | +| [#2428](https://github.com/fixpoint-labs/flow-state-dev/pull/2428) spec FIX-1664 | spec | ~1 (folded through #2434 and FIX-1668) | approval = merge 09-30 16:42Z | design-off ×1 (Codex P1 T7 + Cursor T1: list-then-pick for the task's run; replaced by the task-run link, FIX-1668) · missed-edge-case ×1 (Codex P1: a readable source for harness metrics) · nit ×5 — *2 from 7* | no | — | +| [#2434](https://github.com/fixpoint-labs/flow-state-dev/pull/2434) spec amendment FIX-1664 | spec (amendment) | **0** | **merged 09-30 16:56Z, 5 min after it opened and 6 min before Codex** | spec-ambiguity ×2 (the composer's owner · the hold-resolved-run invariant, both carried into #2436) · missed-edge-case ×2 (Codex P1: clear stale links on a retry · P2: collection identity) · docs-miss ×1 (Codex P2: the evolution record) · nit ×4 — *5 from 9*. **The three Codex threads are unanswered** | — | merge-ready rule (filed) | +| [#2436](https://github.com/fixpoint-labs/flow-state-dev/pull/2436) spec amendment FIX-1664 | spec (amendment) | ~1 | approval = merge 09-30 18:34Z | stale-restatement ×1 (the README table showed one branch) · spec-ambiguity ×1 (Codex P2: the ownerless branch against the goal check) · nit ×3 — *2 from 5*. Two commits align it with FIX-1668's spec, not with review | no | — | +| [#2437](https://github.com/fixpoint-labs/flow-state-dev/pull/2437) spec FIX-1666 | spec | ~1 | approval = merge 09-30 18:33Z | missed-edge-case ×3 (Codex P1: put the approval in the workstream Stream · P2: keep a Deny · SL1: the durable-drain premise, **settled with an in-spec POC, premise held**) · nit ×3 — *3 from 6* | no | — | +| [#2438](https://github.com/fixpoint-labs/flow-state-dev/pull/2438) spec FIX-1667 | spec | ~1 | approval = merge 09-30 17:18Z | missed-edge-case ×3 (Codex P1: run identity across org members · P2 ×2: `.lock` names, case folding) · over-engineered ×1 (SL: use the channel id as is) · nit ×4 — *4 from 8* | no | — | +| [#2439](https://github.com/fixpoint-labs/flow-state-dev/pull/2439) spec amendment FIX-1662 | spec (amendment) | ~1 | approval = merge 09-30 18:32Z | missed-edge-case ×1 (Codex P1: load dispatch-run sessions) · docs-miss ×1 (Codex P2: qualify by visibility) · stale-restatement ×1 (BR-18 against BR-25) · nit ×5 — *3 from 8* | no | — | +| [#2440](https://github.com/fixpoint-labs/flow-state-dev/pull/2440) spec FIX-1668 | spec | ~1 | approval = merge 09-30 18:31Z | missed-edge-case ×3 (Codex P1: the run's owning flow in the link · P2 ×2: publish the change, a failed gate setup) · stale-restatement ×1 (FIX-1664's plan row) · spec-ambiguity ×1 · nit ×3 — *5 from 8* | no | — | +| [#2525](https://github.com/fixpoint-labs/flow-state-dev/pull/2525) spec FIX-1690 | spec | ~1 | approval = merge 10-01 00:24Z, 32 min after it opened | over-engineered ×1 (two durable forms of one line) · nit ×3 — *1 from 4*. **No Codex: usage limit** | no | — | +| [#2592](https://github.com/fixpoint-labs/flow-state-dev/pull/2592) closure plan amendment FIX-1663 | spec (amendment) | ~1 | approval = merge 10-01 18:06Z | stale-restatement ×2 (the body assumed FIX-1664 merged · the amendment told four times) · docs-miss ×1 (EVOLUTION incomplete) · nit ×5 (T5: *"inlining four goal paths will rot when another child goal lands"*) — *3 from 8*. **No Codex: usage limit** | no | — | +| [#2614](https://github.com/fixpoint-labs/flow-state-dev/pull/2614) spec FIX-1723 | spec | ~1 | approval = merge 10-01 19:35Z | spec-ambiguity ×2 (BR-12's waiting count · SL) · over-engineered ×2 (S1 extends `derive.ts` · SL: the switch is FIX-1725's) · missed-edge-case ×1 (SL: the goal Lab carries an org-level seat) · nit ×3 — *5 from 8*. **No Codex** | no | — | +| [#2616](https://github.com/fixpoint-labs/flow-state-dev/pull/2616) spec FIX-1722 | spec | ~1 | approval = merge 10-01 19:34Z | over-engineered ×2 (one copy of FIX-1719's contract · the CoS's tools are FIX-1726's) · docs-miss ×1 (Jump to) · nit ×5 — *3 from 8*. **No Codex** | no | — | +| [#2441](https://github.com/fixpoint-labs/flow-state-dev/pull/2441) impl FIX-1667 | impl | 1 | merge 09-30 18:53Z | missed-edge-case ×2 (Codex P2: filter ownership before the wrapped dispatcher · BR-1 matched any mention) · docs-miss ×2 (the changeset's bump · the draining-board rule) · over-engineered ×1 (POC #2442: the unused wrapper) · nit ×2 — *5 from 7*. Merged head `ae84b66` unread | no | — | +| [#2444](https://github.com/fixpoint-labs/flow-state-dev/pull/2444) impl FIX-1655 | impl | 2 | merge 09-30 20:32Z | missed-edge-case ×3 (**1 *vacuous-assertion***: Codex P1, the goal bypassed the `fsdev ui add` path users run · planning read as warning · the static title icon) · over-engineered ×2 (POC #2446's one token sheet · a second registry walk) · nit ×8 — *5 from 13*. Merged head `82ec64e` unread | no | — | +| [#2445](https://github.com/fixpoint-labs/flow-state-dev/pull/2445) impl FIX-1666 | impl | 2 | merge 09-30 20:04Z | missed-edge-case ×3 (Codex P1: an atomic once-per-feature claim · P2: retry an ask that never reached a person · store reads outside the step) · **design-off ×1 (T6: *"This feels like a core aspect of workforce, why is it in the lab"*; answered by the spec's fence, FIX-1673 filed)** · nit ×2 — *4 from 6*. Merged head `ea475db` unread | **yes** | — | +| [#2447](https://github.com/fixpoint-labs/flow-state-dev/pull/2447) impl FIX-1668 | impl | 1 | merge 09-30 19:59Z | missed-edge-case ×1 (*vacuous-assertion*: Codex P2, the commit-then-throw case threw before the claim-state write) · stale-restatement ×1 (Codex P2: the README API table and the channel-board guide) · over-engineered ×1 · nit ×3 — *3 from 6*. Merged head `20c96ad` unread | no | — | +| [#2449](https://github.com/fixpoint-labs/flow-state-dev/pull/2449) impl FIX-1662 | impl | 1 | merge 09-30 20:23Z | missed-edge-case ×7 (**2 *vacuous-assertion***: Codex P1, the answer leg passed with Approve disabled · Bugbot, the config spelled the held-out slug; and a bearer off loopback · Jump to's declared resources · bare assignee names · panel keys · Bugbot: a returned error left the claim) · over-engineered ×3 (the Lab assembled twice · the start script re-implementing `fsdev dev` · POC #2452's one `openLab`) · philosophy-drift ×1 (SL: a deny-list mirroring the engine's allow-list) · nit ×7 — *11 from 18*. Merged head `7e1afbd` unread | no | — | +| [#2465](https://github.com/fixpoint-labs/flow-state-dev/pull/2465) impl FIX-1664 | impl | 2 | merge 09-30 22:51Z | missed-edge-case ×7 (409 by message regex · a dead Retry · Codex P1: follow the session on a retry · P2 ×3: draw through the item renderers, keep a failed board read, Retry reads the board (+ Bugbot) · Bugbot: the header's seat resolver) · nit ×3 — *7 from 10*. Merged head `103bb9e` unread | no | — | +| [#2527](https://github.com/fixpoint-labs/flow-state-dev/pull/2527) impl FIX-1695, FIX-1684 | impl | 1 | merge 10-01 00:36Z | missed-edge-case ×1 (an exact JSON match on the refusal) · nit ×5 — *1 from 6*. **No Codex, no Bugbot** | no | reviewer exhaustion (filed) | +| [#2529](https://github.com/fixpoint-labs/flow-state-dev/pull/2529) impl FIX-1693, FIX-1694, FIX-1696 | impl | 1 | merge 10-01 00:35Z | missed-edge-case ×1 (any first-read error became a refusal) · nit ×5 — *1 from 6*. **No Codex, no Bugbot** | no | reviewer exhaustion (filed) | +| [#2531](https://github.com/fixpoint-labs/flow-state-dev/pull/2531) impl FIX-1691, FIX-1692 | impl | 1 | merge 10-01 01:24Z | docs-miss ×1 (*overclaim*: "parity with codex/cursor" stamping) · over-engineered ×1 (POC #2534's `itemFields`) · nit ×5 — *2 from 7*. **No Codex, no Bugbot** | no | reviewer exhaustion (filed) | +| [#2532](https://github.com/fixpoint-labs/flow-state-dev/pull/2532) impl FIX-1688, FIX-1689 | impl | 0 (folds on #2537) | merge 10-01 00:34Z, 9 min after it opened | docs-miss ×1 (the `chat-assistant` closure) · nit ×4 — *1 from 5*. **No Codex, no Bugbot** | no | reviewer exhaustion (filed) | +| [#2537](https://github.com/fixpoint-labs/flow-state-dev/pull/2537) impl FIX-1688 follow-up | impl | 0 | merge 10-01 01:57Z | nit ×1 — *0 from 1*. **No Codex, no Bugbot** | no | — | +| [#2538](https://github.com/fixpoint-labs/flow-state-dev/pull/2538) impl FIX-1690 (stop) | impl | 1 | merge 10-01 02:11Z | over-engineered ×1 (reuse `makeBlockingFlow`) · nit ×2 — *1 from 3*. **No Codex, no Bugbot.** Merged head `50f2510` unread | no | reviewer exhaustion (filed) | +| [#2544](https://github.com/fixpoint-labs/flow-state-dev/pull/2544) impl FIX-1690 (door) | impl | 2 | merge 10-01 15:25Z | missed-edge-case ×2 (SL1: a refused line could still be delivered later · a guard that could not fire) · over-engineered ×3 (three helper reuses) · nit ×3 — *5 from 8*. **No Codex (a manual `@codex review` hit the limit), no Bugbot** | no | reviewer exhaustion (filed) | +| [#2565](https://github.com/fixpoint-labs/flow-state-dev/pull/2565) impl FIX-1690 (lab) | impl | 1 | merge 10-01 15:53Z | missed-edge-case ×2 (SL1: Retry after a timeout could double-send · any throw read as not-sent) · over-engineered ×2 (a second send-state machine · three pagination variants) · nit ×3 — *4 from 7*. **No Codex, no Bugbot** | no | reviewer exhaustion (filed) | +| [#2588](https://github.com/fixpoint-labs/flow-state-dev/pull/2588) impl FIX-1649 (rename, boot shift, DevTeam) | impl | 0 | merge 10-01 17:41Z | nit ×5 (all kept, with reasons) — *0 from 5*. **No Codex, no Bugbot** | no | — | +| [#2605](https://github.com/fixpoint-labs/flow-state-dev/pull/2605) impl FIX-1697 | impl | 0 | merge 10-01 19:54Z | nit ×2 — *0 from 2*. **No Codex, no Bugbot** | no | — | +| [#2618](https://github.com/fixpoint-labs/flow-state-dev/pull/2618) impl FIX-1725 | impl | 0 | merge 10-01 19:33Z | nit ×5 — *0 from 5*. **No Codex, no Bugbot** | no | — | +| [#2621](https://github.com/fixpoint-labs/flow-state-dev/pull/2621) impl FIX-1722 | impl | 1 | merge 10-02 01:27Z | missed-edge-case ×6 (Codex P2 ×6: keep the first session id · ids off a secure origin · block the composer until history loads · per-stream failures · resolved suspensions as pending · adopt a newer session) · nit ×4 — *6 from 10*. Merged head `04d7c14` unread | no | — | +| [#2624](https://github.com/fixpoint-labs/flow-state-dev/pull/2624) impl FIX-1723 | impl | 1 | merge 10-02 01:50Z | missed-edge-case ×3 (Codex P1: logical ids for hired seats, the same finding as cycle 19's #2283 · P2 ×2: Retry when asks fail · stop the Lab when Chromium fails) · over-engineered ×1 · nit ×5 — *4 from 9*. Merged head `7781047` unread | no | — | +| [#2641](https://github.com/fixpoint-labs/flow-state-dev/pull/2641) impl FIX-1730–FIX-1734 (closure findings) | impl | 2 | merge 10-02 11:56Z | missed-edge-case ×2 (**both *vacuous-assertion***: a control stub whose signature could drift from `sendTurn` · a planted violation keyed to markers the fix removed) · over-engineered ×3 (POC #2642's one walk · the `--attention` case · `export *` from `send.ts`) · nit ×2 — *5 from 7*. Codex read `2008bbb` with no findings; merged head `bd05070` unread | no | — | +| [#2650](https://github.com/fixpoint-labs/flow-state-dev/pull/2650) impl FIX-1736 | impl | 1 | merge 10-02 17:13Z | missed-edge-case ×2 (weights compared as strings · Codex P2: the mono bold face) · over-engineered ×3 · nit ×1 — *5 from 6*. Merged head `a0ec9f3` unread | no | — | +| [#2654](https://github.com/fixpoint-labs/flow-state-dev/pull/2654) impl FIX-1735 | impl | 1 | merge 10-02 17:28Z | missed-edge-case ×1 (Codex P2: a record absent at the first read stayed cached) · over-engineered ×3 (three shared helpers) · nit ×2 — *4 from 6*. Merged head `bc16d39` unread; two more FIX-1735 fixes landed later on #2678 | no | — | +| [#2669](https://github.com/fixpoint-labs/flow-state-dev/pull/2669) impl FIX-1747 | impl | 1 | merge 10-03 01:54Z | over-engineered ×1 (a whole-page negative match) · *1 from 1*. Codex clean on `59dc49d` (manual); merged head `5fefdde`, a merge of `main`, unread | no | — | +| [#2671](https://github.com/fixpoint-labs/flow-state-dev/pull/2671) impl FIX-1749 | impl | 1 | merge 10-02 20:39Z | missed-edge-case ×1 (Codex P2: the setup-failure write) · docs-miss ×2 (the changeset, an *overclaim* · the ordering contract in `items.md`) — *3 from 3*. Merged head `0cc4206` unread | no | — | +| [#2675](https://github.com/fixpoint-labs/flow-state-dev/pull/2675) impl FIX-1737 A | impl | 1 | merge 10-04 00:01Z, 26 h after its round | missed-edge-case ×3 (a re-read that graded a different snapshot · Codex P2 ×2: duplicate short names in mentions · delivery state after a mention) · over-engineered ×2 (per-element styles · a second composer) · docs-miss ×1 (the `goals/lib` table) · nit ×1 — *6 from 7*. Merged head `839a5f5`, a merge of `main`, unread | no | — | +| [#2681](https://github.com/fixpoint-labs/flow-state-dev/pull/2681) impl FIX-1737 B | impl | 1 | merge 10-04 00:04Z | missed-edge-case ×4 (**1 *vacuous-assertion***: Codex P2, rows with `min: 0` never exercised · in-flight sends · asks from unrelated flows · an `endsWith` match) · over-engineered ×1 (one needs-then-running rule) · nit ×2 — *5 from 7*. Merged head `a557e97` unread | no | — | +| [#2682](https://github.com/fixpoint-labs/flow-state-dev/pull/2682) impl FIX-1737 C | impl | 1 | merge 10-04 00:32Z | missed-edge-case ×3 (an unfiltered walk of every session · Codex P2 ×2: refresh asks on suspend, Escape where advertised) · over-engineered ×1 · nit ×3 — *4 from 7*. Merged head `2c794e3` unread | no | — | +| [#2683](https://github.com/fixpoint-labs/flow-state-dev/pull/2683) impl FIX-1737 D | impl | 1 | merge 10-04 00:49Z | missed-edge-case ×4 (**1 *vacuous-assertion***: Codex P2, the expected wait id was built from the page under test · a span count · Codex P2 ×2: truncated reads, distinct running sessions) · over-engineered ×1 · nit ×8 — *5 from 13*. Merged head `c10339a` unread | no | — | +| [#2709](https://github.com/fixpoint-labs/flow-state-dev/pull/2709) closure FIX-1663 | impl (closure) | 1 | merge 10-04 10:56Z | missed-edge-case ×5 (**3 *vacuous-assertion***: Codex P1 T7, `GOAL_PART=bogus` printed PASS with 0 verdicts · P2 T9, b0 passed with no `GUESSED` line · P2 T10, `GOAL_PART=b` skipped DevForce's org check; and Queued rows (T4+T8) · P2 T11, a narrowed rerun graded the full list) · over-engineered ×1 (T1+T2: a second copy of `goals/lib/shift-manager.mts`, folded from POC #2711) · nit ×3 — *6 from 9*. **No Bugbot (usage limit)**; Codex read `49fd744`, merged head `5d5358b` unread | no | — | + +**Load.** 29 fix waves across the 29 implementation artifacts (cycle 20: 57 across 15), and 104 +non-`nit` findings, 3.6 each (cycle 20: 8.7). The busiest is #2449 at 11. No PR took more than +two waves, and five took none. Eleven of the 104 are *vacuous-assertion*, or 11% (cycle 20: 4%; +cycle 19: 10%). Three of the eleven are in the closure runner. The eighteen direction artifacts carry 62 non-`nit` +findings, 3.4 each (cycle 20: 8.0). None reached a second round. Stale-restatement is 11 of the 62 +(cycle 20: 5 of 48), and #2619 carries 4 of them, all from the rename. Across all 166 non-`nit` +findings: missed-edge-case is 89 (54%), over-engineered 40, stale-restatement 12, docs-miss 11, +spec-ambiguity 10, design-off 3 and philosophy-drift 1. + +**Claims (looped / settled / verdicts):** 0 / 0 on every artifact. No `settle-claim` ran. #2437's +durable-drain premise was settled by an in-spec POC (`poc/durable-drain/`) on the first round, +before anyone argued it. It is not a loop, so it is not counted. + +## The closure's runs + +| Run | On | What it found | +|---|---|---| +| First QA pass | `main`, 09-30 | FIX-1688 to FIX-1696 (nine), FIX-1697. FIX-1688: the theme swap that FIX-1662's stylesheet deferred to *"when FIX-1655's package lands"* never happened | +| Closure run 1 | `c29aafa4f` (10-02) | FIX-1730 (b0 guessed 3 steps), FIX-1731 (54 literal colours), **FIX-1732, FIX-1733 and FIX-1734 (three controls WRONG)**, FIX-1735 (observed) | +| Closure run 2 | `9bbc77b04` on `48c31cfa6` | a4 refused by the door while the harness started (FIX-1735). All six controls RIGHT, part 3 PASS | +| — | 10-02 12:16Z | FIX-1736 (the fonts never load) and FIX-1737 (v2's look), filed outside a run row. The look goal read the computed `font-family` string, which names the family whether or not it loaded | +| Closure run 3 | `3537738a7` on `a01a0b7bd` (10-04) | **Four findings from merges to `main`**: the project level FIX-1718 shipped, `stream-asks` renamed `feed-ask` (FIX-1737 C), eng.em's four channels, two writes ER-15 does not name. Part 3: three child checks broken by FIX-1737 B (`NavIcon` fill), FIX-1737 D (Queued rows hidden) and FIX-1718 (a hardcoded member list) | +| Runs 4 and 5 | `dd9054581`, `25d026528` | a1's open workstream case, decided; ER-15's room writes pinned. PASS on `25d026528` | +| Run 6 | `01444863c` | PASS after Codex's five findings and POC #2711. Negatives: `GOAL_PART=bogus` now exits 1 | + +## The class: controls that break when the closure reuses them + +Run 1 graded each control at the signal it names, and three came back WRONG: + +| Control (owner) | What went wrong | Cause | Cycle 19's twin | +|---|---|---|---| +| `static-names` (FIX-1662) | reddened a4 as well as b:teams | it wrote seats in as `{ id, kind }`, which drops the `door` FIX-1690 added later | FIX-1624: *no-landing broke more legs than the landing* | +| `optimistic-post` (FIX-1662) | b:post red, a4 green | it swaps `transcript.ts`; since FIX-1690, an `@worker` line goes through `send.ts` | FIX-1625: *controls stopped failing once FIX-1609 added a live stream route* | +| `worker-session` (FIX-1664) | a2 green | with no other session on the seat's flow it fell back to the run session, and DevForce's coder has none | FIX-1623: *the scripted answer came too fast for no-live to fail* | + +The three shapes are one for one: a control breaks **more than its signal**, a control **misses a +second path** to the outcome it removes, or the fixture leaves it **nothing to change**. Each +control was RIGHT at its owning check, on its owning Lab. Each one went WRONG when the closure +applied it to another Lab and to legs that the owning check never ran. + +**Why nothing caught it before the closure.** +- **The closure plan said what each control would do on its legs, and could not run it yet.** + On #2426, Codex T9 found that `optimistic-post` could not isolate a4. The fold answered *"it + removes the send on both trees, so it's now graded as a pair"*. That was true on 09-30. FIX-1690's + spec (#2525) opened that night, and its lab PR (#2565) added the `send.ts` path and the doors on + 10-01. That is after the plan, and both of those PRs had no correctness-bot pass. +- **The owning checks could not see it.** `it-opens-a-lab` grades `optimistic-post` at *"the post + is in the stored transcript"*. That signal stayed red after FIX-1690. A run of the child check and + its controls on #2565 would have shown RIGHT. +- **The closure is where reuse is graded.** Run 1 found all three, and run 2 confirmed all six + controls RIGHT on the next `main`. That cost one fix PR (#2641, two waves) and one re-run. + +## Proposal skipped + +The epic's review cost was low on every measure the ledger tracks. The rework it is remembered for +was found by closure runs, on the path that `orchestration.md` → *The closure issue* gives them. +That path says the closure *"is the most expensive child and the only one that repeats. That cost +is the QA."* Each candidate below either has no home that is cheaper than the run that found it, or +is covered already and was caught on the first read. + +## Candidates considered and dropped + +- **Controls that break when reused (the class above).** The recurrence is real: two consecutive + closures, three controls each, the same three shapes. But the earliest point that can see each + one is the closure's first control run. Two of the three were caused by a sibling that merged + *after* the closure plan was written. A rule that asks the plan's author to predict later + siblings would add a checklist and catch nothing. Run 1 is the calibration run, and it worked. + Carried as claim 1. +- **A selection that runs nothing reads as PASS** (#2709 T7, T9, T10; cycle 19's #2348 row 2). This + is covered by BP-003 (*a command that never ran the check*) and by `issue-implement`'s *the unit + is the assertion* rule (cycle 17's fix A). Codex found all three on the first head, and one wave + fixed them. A guard in the verdict protocol is filed. +- **Parallel slices verified alone and broken together.** FIX-1737's four slices each ran the + sibling checks they touched, on their own heads. They merged within 48 minutes, and the combined + tree first ran at closure run 3, which found the three broken child checks. Closure item 3 exists + for this: *"they catch a later merge breaking an earlier child"*. It caught all three. #2592 T5 had + also predicted that the closure's inlined goal paths would rot. +- **A check that graded a proxy** (FIX-1736: the computed `font-family` string; the look goals that + graded token values while v2's look was not drawn). This is *vacuous-assertion*. Tenet 7, BP-003 + and `goals/README.md` → *Proving a check can fail* already name it. It was found by eye and filed + as two issues. That is the rule applied late, not a gap in the rule. +- **A stale DevTeam store across reruns** (`3537738a7`: each DevTeam start now gets a fresh store, + never the checkout's `devteam.sqlite`). One instance, fixed in the goal. + +## Filed, not proposed + +- **Reviewer exhaustion grew from 3 artifacts to 16.** Codex hit its limit at 10-01 00:08Z. The + first PR after that with a Codex review is #2621, opened at 19:56Z. Bugbot reports a usage limit on every PR from #2529 + (10-01 00:14Z) through #2709 (10-04). Eleven implementation PRs and five direction artifacts got no + correctness-bot pass on any head. FIX-1690's three PRs are among them. Two closure findings sit in + code from that window: FIX-1733's second path (#2565) and FIX-1735's door refusal (#2544). Whether + a Codex pass would have found them is not derivable. **The owner's decision from cycle 20 is still + open: buy capacity, or name which automation counts as a correctness pass.** +- **The closure fixed run 3's findings in its own branch.** `orchestration.md` says every finding + becomes a child of the epic and that a run with findings opens no PR. Run 1 did this (FIX-1730 to + FIX-1735). Ten commits in the closure branch answer runs 3 and 4 instead. One of them is a + product fix (`1d1e1cb`, the sidebar icons' fill), and one closes a case the specs left open + (a1's workstream set, `6c6030939`). The re-run discipline held, because each fix was re-run on one + commit. The audit trail did not, because none of those findings has an issue. +- **#2434 merged 5 minutes after it opened**, before Codex. Codex posted three threads 6 minutes + after the merge. One is a P1 (clear stale run links on a retry). None is answered. FIX-1668's spec + later took the run link, so the P1 may be moot. The PR does not say so. +- **Four children never blocked the closure.** FIX-1671, FIX-1673 and FIX-1675 (features) and + FIX-1705 (a bug: a coding run that outlasts a stop is charged an abandonment) are children of + FIX-1649 with only `related` links to FIX-1663. The epic wrapped with all four open. The rule is + that every child blocks the closure, including late ones. +- **Linear lags the merges.** FIX-1663, FIX-1722, FIX-1723, FIX-1737 and FIX-1749 read *In + Development*. FIX-1697, FIX-1725 and FIX-1730 to FIX-1736 read *In Review*. Their PRs have + merged. +- **POC #2642 is still open** after #2641 adopted it (`bd05070`). +- **FIX-1735 continued on another epic's PR.** `1f2dadd45` (*a mid-run request-state write keeps the + persisted items*) and `f23ae7c78` landed on #2678 (FIX-1719) after #2654 merged. +- **A zero-check guard in `goals/lib/verdict.mts`.** `runGoal` passes when `failures` is empty, so a + runner that selects nothing passes unless it adds its own check, as #2709 now does. A required + count of checks run, with FAIL at zero, would make that the default for every runner. It would not + catch T9 or T10, which ran some checks and skipped others. + +## Scoring the previous cycles' fixes and claims + +``` +46b20ac (cycle 20's race-primitive trigger, on main 10-01 17:12Z): + #2621 first head 6343623 CARRIES · #2654 first head 7c9e932 CARRIES + every branch opened after 17:12Z carries it; #2445, #2449, #2465, #2544, #2565 forked before it +d50a3a6 (cycle 19's control-unit convergence, on main 09-29 12:00Z): + #2426 (opened 09-30) CARRIES · #2709 first head 5446d09 CARRIES +``` + +- **Cycle 20's claim 1 (does the primitive trigger cut race tails?): no qualifying instance.** No + PR that carries the trigger had two interleaving findings on one structure. The nearest is a + pre-fix PR. #2445 drew three findings on `raiseAsk`'s once-per-feature claim (T3, T4 and #2449's + Bugbot T16). Two waves closed them with the existing store operations (*"No new API was + needed"*), which is the trigger's own carve-out. Hold. +- **Cycle 20's claim 2 (reviewer exhaustion): 16, up from 3.** See *Filed*. +- **Cycle 20's claim 3 (changeset length): 0, unexplained.** Eleven PRs added fragments. Seven of + them got a Codex pass, and none drew the one-sentence P1 (cycle 20: 7 of 15). No guard landed. + Carry once more. +- **Cycle 19's claim 1 (does converging the control unit move the plan and the first head?): + yes, for one unconfounded instance.** Codex read #2709's first reviewed head (`49fd744`) and raised no + control-unit finding (cycle 19's #2348: row 1). Its three *pass on an absence* findings are + cycle 19's row 2 again, which is a different claim (see *Candidates considered and dropped*). The closure plan's control table + names signals (*"a4 and b's post step fail together"*). The runner printed WRONG at the named + signal on the first run, and that is how the three controls in *The class* became visible. Close. +- **Cycle 19's claim 2 (do folds after the last pass cost anything?): not derivable, and the count + rose again.** 25 of 29 implementation PRs merged a head that no bot read (cycle 20: 10 of 15). + Codex read the merged head of no implementation PR that had a fix wave. Carry. +- **Cycle 18's claim 3 (a worker that cannot dispatch): 0 instances.** Hold. + +## Claims to test next cycle + +1. **Do reused controls keep breaking in the same three shapes on the closure's first run?** + Baseline: cycle 19 had 3, and cycle 21 had 3 of the 5 non-baseline closure controls. Record each + WRONG control by shape (more than its signal · a missed second path · nothing to change), and say + whether the change that broke it merged after the closure plan. A third closure with the same + result, from controls whose breaking change merged *before* the plan, would mean the plan could + have seen it. That would earn a line in the closure plan's control table. +2. **Reviewer exhaustion.** Count artifacts with no correctness-bot pass on any head. Baseline: 16. +3. **Unread merged heads.** Baseline: 25 of 29. +4. **Changeset length**, once more. Baseline: 0 of 7 fragments that Codex read. + +## Finding map + +`T`-numbers are positions in each PR's review-thread list, oldest first. `+` joins threads collapsed +into one finding. *vac* is *vacuous-assertion*. *SL* is an item of a second-look body. *mee* is +missed-edge-case. + +- **#2421** design-off ×1: SL1. spec-ambiguity ×3: SL2 · T1+T7 · T4. over-engineered ×1: SL3. nit ×4: T2 · T3 · T5 · T6. +- **#2423** stale-restatement ×1: T2. nit ×1: T1. +- **#2619** stale-restatement ×4: T1 · T2 · T4 · T5. nit ×2: T3 · T6. +- **#2648** mee ×4: T7 (vac) · T8 · T9 · T10. over-engineered ×1: T2. nit ×5: T1 · T3–T6. +- **#2424** mee ×3: T1 · T2 · T3. over-engineered ×1: T4+SL. stale-restatement ×1: T7. nit ×3: T5 · T6 · T8. +- **#2425** mee ×3: T10 · T11 · T13. spec-ambiguity ×1: T12. nit ×9: T1–T9. +- **#2426** mee ×2: T9 · SL (D2). nit ×8: T1–T8. +- **#2428** design-off ×1: T1+T7. mee ×1: T8. nit ×5: T2–T6. +- **#2434** spec-ambiguity ×2: T1 · T5. mee ×2: T7 · T8. docs-miss ×1: T9. nit ×4: T2 · T3 · T4 · T6. +- **#2436** stale-restatement ×1: T4. spec-ambiguity ×1: T5. nit ×3: T1–T3. +- **#2437** mee ×3: T4 · T5 · SL1. nit ×3: T1–T3. +- **#2438** mee ×3: T5 · T6 · T7. over-engineered ×1: SL. nit ×4: T1–T4. +- **#2439** mee ×1: T7. docs-miss ×1: T8. stale-restatement ×1: T2. nit ×5: T1 · T3–T6. +- **#2440** mee ×3: T6 · T7 · T8. stale-restatement ×1: T3. spec-ambiguity ×1: T4. nit ×3: T1 · T2 · T5. +- **#2525** over-engineered ×1: T1. nit ×3: T2–T4. +- **#2592** stale-restatement ×2: T1 · T2. docs-miss ×1: T3. nit ×5: T4–T8. +- **#2614** spec-ambiguity ×2: T1 · SL. over-engineered ×2: T2 · SL. mee ×1: SL. nit ×3: T3–T5. +- **#2616** over-engineered ×2: T2 · the FIX-1726 scope fold. docs-miss ×1: T7. nit ×5: T1 · T3–T6. +- **#2441** mee ×2: T2 · T3. docs-miss ×2: T1 · T6. over-engineered ×1: T7. nit ×2: T4 · T5. +- **#2444** mee ×3: T10 (vac) · T11 · T12. over-engineered ×2: T7 · T13. nit ×8: T1–T6 · T8 · T9. +- **#2445** mee ×3: T3 · T4 · T5. design-off ×1: T6. nit ×2: T1 · T2. +- **#2447** mee ×1: T6 (vac). stale-restatement ×1: T5. over-engineered ×1: T4. nit ×3: T1–T3. +- **#2449** mee ×7: T11 (vac) · T12 · T13 · T14 · T15 · T16 · T17 (vac). over-engineered ×3: T2 · T4 · T18. philosophy-drift ×1: T1. nit ×7: T3 · T5–T10. +- **#2465** mee ×7: T1 · T3 · T6 · T7 · T8 · T9+T10 · T11. nit ×3: T2 · T4 · T5. +- **#2527** mee ×1: T1. nit ×5: T2–T6. +- **#2529** mee ×1: T2. nit ×5: T1 · T3–T6. +- **#2531** docs-miss ×1: T3. over-engineered ×1: T7. nit ×5: T1 · T2 · T4–T6. +- **#2532** docs-miss ×1: T4. nit ×4: T1–T3 · T5. +- **#2537** nit ×1: T1. +- **#2538** over-engineered ×1: T1. nit ×2: T2 · T3. +- **#2544** mee ×2: T1 · SL1. over-engineered ×3: T2 · T4 · T5. nit ×3: T3 · T6 · T7. +- **#2565** mee ×2: T3 · SL1. over-engineered ×2: T1 · T2. nit ×3: T4–T6. +- **#2588** nit ×5: T1–T5. +- **#2605** nit ×2: T1 · T2. +- **#2618** nit ×5: T1–T5. +- **#2621** mee ×6: T5–T10. nit ×4: T1–T4. +- **#2624** mee ×3: T7 · T8 · T9. over-engineered ×1: T3. nit ×5: T1 · T2 · T4–T6. +- **#2641** mee ×2: T1 (vac) · T4 (vac). over-engineered ×3: T5 · T6 · T7. nit ×2: T2 · T3. +- **#2650** mee ×2: T3 · T6. over-engineered ×3: T2 · T4 · T5. nit ×1: T1. +- **#2654** mee ×1: T1. over-engineered ×3: T2 · T3 · T5. nit ×2: T4 · T6. +- **#2669** over-engineered ×1: T1. +- **#2671** mee ×1: T1. docs-miss ×2: T2 · T3. +- **#2675** mee ×3: T2 · T6 · T7. over-engineered ×2: T1 · T4. docs-miss ×1: T5. nit ×1: T3. +- **#2681** mee ×4: T5 · T6 · T7 (vac) · T8. over-engineered ×1: T1+T2. nit ×2: T3 · T4. +- **#2682** mee ×3: T1 · T6 · T7. over-engineered ×1: T5. nit ×3: T2–T4. +- **#2683** mee ×4: T4 · T11 · T12 · T13 (vac). over-engineered ×1: T7. nit ×8: T1–T3 · T5 · T6 · T8–T10. +- **#2709** mee ×5: T4+T8 · T7 (vac) · T9 (vac) · T10 (vac) · T11. over-engineered ×1: T1+T2. nit ×3: T3 · T5 · T6. + +**Implementation non-`nit`:** 5 + 5 + 4 + 3 + 11 + 7 + 1 + 1 + 2 + 1 + 0 + 1 + 5 + 4 + 0 + 0 + 0 + 6 ++ 4 + 5 + 5 + 4 + 1 + 3 + 6 + 5 + 4 + 5 + 6 = **104** (in table order), of which *vacuous-assertion* +11. **Direction non-`nit`:** 5 + 1 + 4 + 5 + 5 + 4 + 2 + 2 + 5 + 2 + 3 + 4 + 3 + 5 + 1 + 3 + 5 + 3 = +**62** (in table order), of which stale-restatement 11 and *vacuous-assertion* 1.