Skip to content

fix: drop dead unreadable verdict from ratification help and defer row - #14253

Open
loom-fleet-dispatch[bot] wants to merge 5 commits into
mainfrom
feature/issue-14125
Open

loom-fleet-dispatch[bot] wants to merge 5 commits into
mainfrom
feature/issue-14125

Conversation

@loom-fleet-dispatch

@loom-fleet-dispatch loom-fleet-dispatch Bot commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

Summary

After #14070 the champion-hold ratification verb exits 3 on any unreadable candidate before scan runs, so verdict=unreadable is never printed. This removes the stale references to it; no behaviour change.

Changes

  • loom-daemon/src/cli/champion_hold_cmd.rs: --help for Ratification lists only verdict=<clear|hold> (the action= list is unchanged); the dead Verdict::Unreadable arm in ratification_line becomes fail("head content") (a diverging exit-3 refusal, not a panic) with a comment pointing at the gate. The enum variant stays (the role-runner lane uses it).
  • defaults/.claude/commands/loom/champion-ratification-hold.md: the defer row no longer says "no hold standing", the "Fail closed" bullet now says an unreadable candidate always defers, and the doc notes that a read-failure tick does not re-assert loom:blocked or the park record.

Acceptance Criteria Verification

  • --help enumerates only emittable verdicts (clear|hold); action= list unchanged. Verified in the diff of the doc comment (clap help source).
  • Dead arm replaced with a diverging fail("head content") (exit 3) carrying a comment pointing at the gate; Verdict::Unreadable variant untouched.
  • defer row no longer says "no hold standing"; the optional clause is included.
  • No behaviour change: the gate at unreadable_candidates(...) already calls fail() before scan, so the arm cannot be reached.

Test Plan

TDD: N/A. Help text, a dead-arm cleanup and a doc-wording change; no behaviour change to pin with a new test.

  • cargo test -j2 -p loom-daemon champion_hold: passed (exit 0).
  • cargo fmt --check -p loom-daemon: clean.
  • CI's cargo clippy --package loom-daemon --all-targets -- -D warnings ... line, run verbatim: passed. It had to run with RUSTC_WRAPPER="" for this one command, because the host's shared sccache server was failing with a missing /tmp/loom-nt/<hash> temp dir. I didn't touch the shared server or any host tool.
  • loom-daemon preflight --issue 14125: "no enabled buildGate — skipped".
  • Not run: defaults/scripts/tests/test-champion-ratification-hold.sh. A grep shows it doesn't reference the old wording or unreadable, so I left it to CI.
  • CI was still running at hand-off.

Closes #14125

🤖 Generated with Claude Code

loom dashboard

…w wording

The ratification verb exits 3 on any unreadable candidate before scan
runs, so verdict=unreadable is never printed. Trim it from --help, make
the dead Verdict::Unreadable arm an explicit unreachable! pointing at the
gate, and drop the stale 'no hold standing' precondition from the defer
row.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@loom-fleet-dispatch loom-fleet-dispatch Bot added the loom:review-requested PR ready for Judge to review. Applied by: Builder when opening PR. label Nov 18, 2026
@loom-fleet-dispatch

loom-fleet-dispatch Bot commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor Author

PR #14253's loom:reviewing claim is held by the judge of sweep step-judge-14253-1795006926 on host host-7ec0fa4c. This comment is a PR lease record (#9401): its liveness signal is its own forge-assigned updated_at, renewed while the judge works and released when it is done. Another judge that finds this lease live stands down before touching a worktree. See defaults/docs/lease-record.md § "PR-phase leases".

@loom-fleet-dispatch loom-fleet-dispatch Bot added the loom:reviewing Judge is reviewing this PR. Applied by: Judge only. Stale after LOOM_STALE_REVIEWING_MINUTES (30m). label Nov 18, 2026
@loom-fleet-dispatch

Copy link
Copy Markdown
Contributor Author

✅ Approved!

A 7-line, no-behaviour-change tidy that matches its issue exactly. I re-derived the central claim rather than taking the PR's word for it.

The unreachability claim holds (measured, by reading the code)

scan (loom-daemon/src/champion_hold/ratification.rs:167-185) returns Verdict::Unreadable(unreadable) only when hits.is_empty() && !unreadable.is_empty(), and it pushes into unreadable under exactly f.status != "removed" && is_candidate(f, cfg) && f.head_content.is_none(). That is token-for-token the filter in unreadable_candidates (ratification.rs:195-201), so !unreadable.is_empty() ⟺ !unreadable_candidates(files, cfg).is_empty().

ratification_line (cli/champion_hold_cmd.rs:283-312) runs that gate before scan, with the same &files / &cfg bindings, and fail is let fail = |what: &str| -> ! { unreadable("ratification", t.pr, what) } (champion_hold_cmd.rs:235) — a diverging closure, so exit 3 really does precede scan. Nothing between the gate and scan mutates files (only the fetch_trusted_comments read, itself unwrap_or_else(|| fail(...))). So the replaced arm is genuinely unreachable, and the unreachable! is a panic-on-contract-violation rather than a swallowed state — correct, and unreachable! is checked in release builds (no UB).

The enum variant correctly stays

Verdict::Unreadable is live at role_runner/ratification_lane.rs:208 (the lane calls scan directly, with no gate in front, so it does observe the variant) and in decide at ratification.rs:369. Removing it would have broken that lane; the PR's choice matches the Curator AC.

Spec reconciliation

spec=#14125 (loom:curated, loom:building, open) comments=5, all loom-fleet-dispatch[bot] (two lease records, Curator enhancement, Champion promotion, lease yield); no human comment and no body edit since REVIEW_STARTED. No earlier spec baseline available for this PR; reviewed against the current requirements. No material difference between the issue's four ACs and the PR's scope:

  • AC1 --help lists only verdict=<clear|hold>, action= untouched — ✅ in the diff.
  • AC2 dead arm → unreachable!("gated by unreadable_candidates above") with a comment pointing at the gate, enum variant untouched — ✅.
  • AC3 defer row no longer says "no hold standing", plus the optional loom:blocked/park-record clause — ✅.
  • AC4 no behaviour change — ✅ per the proof above.

The issue's "and its installed .claude/ copy" is a no-op here: git ls-files shows defaults/.claude/commands/loom/champion-ratification-hold.md is the only tracked copy (the installed one is refreshed by resync-installed.sh), and the Curator explicitly scoped it out.

Reworded defer row is accurate

The new clause ("a read-failure tick does not re-assert loom:blocked or the park record") matches decide: Action::Defer is produced only by (Verdict::Unreadable(_), Episode::None | Episode::Released { .. }), and the Stands arm — which is what re-asserts the hold — is a different arm. The surviving defer) echo at line 91 of the same file ("unreadable or inconclusive") is still accurate and rightly untouched.

Tests

  • CI: LOOM-CHECKS-GREEN a6062c9 (exact reviewed head), merge state CLEAN. I did not re-run the suite locally — CI is the test gate (Host-side test burden to a bare minimum: CI is the test gate, Builders don't run the suite (operator direction 2026-10-13) #11690) and my one specific doubt was the unreachability, which I settled by reading the code above.
  • TDD: N/A accepted as plausible: a doc comment, a dead-arm replacement and a doc-wording change, with no behaviour to pin. Advisory only.
  • The PR's claim that defaults/scripts/tests/test-champion-ratification-hold.sh does not depend on the old wording checks out — I grepped it: its only defer reference is tick defer 0; check "" "$WRITES" ... at line 102, a behavioural assertion, with no "no hold standing" or unreadable string.
  • Review-feedback gate: REVIEW_FEEDBACK_STATE=CLEAR (0 reviews, 0 inline threads, resolution source graphql).
  • Note on the builder's RUSTC_WRAPPER="" workaround: that is a per-command env override for a broken shared sccache temp dir, not a change to any host tool — correct handling, and CI's own clippy line passed independently.

One tracked follow-up (non-blocking): the same staleness survives in action=

Filed #14271. The gate that kills verdict=unreadable also kills action=defer on this verb's stdout, for the identical reason: Action::Defer comes only from (Verdict::Unreadable(_), …) in decide, and ratification_line's verdict is post-gate, so it is only ever Clear or Hold. grep -rn "LOOM-CHAMPION-RATIFICATION" confirms ratification_line:308 is the sole producer of that line, so there is no other path that could emit it. The help comment therefore still advertises action=<…|defer>.

This is not a defect in this PR — issue #14125 and the Curator's AC1 both pinned "action= list unchanged", and Champion promoted it on that scope. I am not re-litigating a ratified scope over one token; #14271 carries it with the measurement. The tick table needs nothing: it derives defer from the exit code, not from action=.

loom dashboard

@loom-fleet-dispatch loom-fleet-dispatch Bot added loom:pr PR approved by Judge, ready for Champion auto-merge. Applied by: Judge. and removed loom:review-requested PR ready for Judge to review. Applied by: Builder when opening PR. loom:reviewing Judge is reviewing this PR. Applied by: Judge only. Stale after LOOM_STALE_REVIEWING_MINUTES (30m). labels Nov 18, 2026
@loom-fleet-dispatch

loom-fleet-dispatch Bot commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor Author

PR #14253's (no label: the lease is the claim) claim is held by the champion of sweep step-champion_merge-14253-1795024185 on host host-d0749b00. This comment is a PR lease record (#9401): its liveness signal is its own forge-assigned updated_at, renewed while the champion works and released when it is done. Another champion that finds this lease live stands down before touching a worktree. See defaults/docs/lease-record.md § "PR-phase leases".

@loom-fleet-dispatch

Copy link
Copy Markdown
Contributor Author

Champion: Holding for Human Merge

Judge-approved and passing the mechanical safety criteria, but not
auto-merging:

  • Blast radius: kept set, merge gate itself: loom-daemon/src/cli/champion_hold_cmd.rs (the champion-hold ratification verb that decides the ratification hold) and champion-ratification-hold.md (its tick table) decide whether a PR merges; the new unreachable! arm would panic the gate verb if the unreadable_candidates gate ever stops preceding scan. The gate cannot vouch for a change to itself.

Next steps — this hold stays in force until one of these happens; Champion re-reading the same diff will not clear it:

Keeping loom:pr. This PR stays in the queue and is re-checked each tick against the release conditions above.


Automated by Champion role

loom dashboard

@loom-fleet-dispatch loom-fleet-dispatch Bot added the loom:blocked Waiting on something named (another change, a capability); released once it resolves label Nov 18, 2026
@rjwalters

Copy link
Copy Markdown
Member

Operator ruling (rjwalters, 2026-11-19, attended /fleet:decide round 20261119-r17, q49): back to Doctor. Recorded by the attended agent session.

Two things to fix before this returns for review:

  1. It conflicts in intent with fix: drop unreachable action=defer from champion-hold ratification --help #14322 (released for merge in the same ruling): fix: drop unreachable action=defer from champion-hold ratification --help #14322 drops the unreachable defer from the action=<…> help line in champion_hold_cmd.rs; this PR keeps it. Rebase on main and settle one wording.
  2. Verdict::Unreadable(_) => unreachable!(…) turns a case the comment says cannot happen into a panic of the champion-hold command. Prefer a non-panicking arm (return an error / exit 3 with a message) so a future change to unreadable_candidates degrades to a reported refusal rather than a crash.

Labels: loom:changes-requested; the Champion hold stays until a fresh Judge approval.

@rjwalters rjwalters added loom:changes-requested PR requires changes before re-review (Judge requested modifications). Applied by: Judge. and removed loom:pr PR approved by Judge, ready for Champion auto-merge. Applied by: Judge. labels Nov 19, 2026
@loom-fleet-dispatch

loom-fleet-dispatch Bot commented Nov 19, 2026 •

Copy link
Copy Markdown
Contributor Author

PR #14253's loom:treating claim is held by the doctor of sweep step-doctor-14253-1795128609 on host host-a6032f0c. This comment is a PR lease record (#9401): its liveness signal is its own forge-assigned updated_at, renewed while the doctor works and released when it is done. Another doctor that finds this lease live stands down before touching a worktree. See defaults/docs/lease-record.md § "PR-phase leases".

@loom-fleet-dispatch loom-fleet-dispatch Bot added the loom:treating Doctor is fixing this bug or PR. Applied by: Doctor. Stale after LOOM_STALE_TREATING_MINUTES (60m). label Nov 19, 2026
# Conflicts:
#	loom-daemon/src/cli/champion_hold_cmd.rs
@loom-fleet-dispatch loom-fleet-dispatch Bot added loom:review-requested PR ready for Judge to review. Applied by: Builder when opening PR. and removed loom:changes-requested PR requires changes before re-review (Judge requested modifications). Applied by: Judge. loom:treating Doctor is fixing this bug or PR. Applied by: Doctor. Stale after LOOM_STALE_TREATING_MINUTES (60m). labels Nov 19, 2026
@loom-fleet-dispatch

Copy link
Copy Markdown
Contributor Author

🩺 Doctor handoff — per-finding status at 48753b2:

  • Operator ruling, item 1 (conflicts in intent with fix: drop unreachable action=defer from champion-hold ratification --help #14322, settle the action= wording): fixed — merged origin/main (which carries fix: drop unreachable action=defer from champion-hold ratification --help #14322) and resolved loom-daemon/src/cli/champion_hold_cmd.rs:61-62 to main's action=<none|hold|stands|release|clear> (no defer) while keeping this PR's verdict=<clear|hold> narrowing. One wording now stands repo-wide.
  • Operator ruling, item 2 (no panic in the merge gate): fixed at loom-daemon/src/cli/champion_hold_cmd.rs:303 — Verdict::Unreadable(_) now degrades to fail("head content") (the same reported refusal as the unreadable_candidates gate: stderr message + exit 3, which the tick table maps to defer) instead of unreachable!.

Class check: no other unreachable!/panic instance of this class in the PR's changed files (the only remaining panic! is a "not needed" test stub at line 544).

Verified locally: cargo check -p loom-daemon, the 3 champion_hold_cmd unit tests, cargo clippy -p loom-daemon --all-targets (clean), and defaults/scripts/tests/test-champion-ratification-hold.sh (31/31). The merge-conflict resolution itself is more than marker shuffling (help-text wording settled), so this is not marked conflict-only.

CI on 48753b2 was still running at hand-off (snapshot: pending). loom:blocked (Champion hold) left intact per the ruling — fresh Judge approval governs.

Ready for re-review.

loom dashboard

@loom-fleet-dispatch

loom-fleet-dispatch Bot commented Nov 19, 2026 •

Copy link
Copy Markdown
Contributor Author

PR #14253's loom:reviewing claim is held by the judge of sweep step-judge-14253-1795130162 on host host-d0749b00. This comment is a PR lease record (#9401): its liveness signal is its own forge-assigned updated_at, renewed while the judge works and released when it is done. Another judge that finds this lease live stands down before touching a worktree. See defaults/docs/lease-record.md § "PR-phase leases".

@loom-fleet-dispatch loom-fleet-dispatch Bot added the loom:reviewing Judge is reviewing this PR. Applied by: Judge only. Stale after LOOM_STALE_REVIEWING_MINUTES (30m). label Nov 19, 2026
@loom-fleet-dispatch

Copy link
Copy Markdown
Contributor Author

Evaluating this PR per user request.

@loom-fleet-dispatch

Copy link
Copy Markdown
Contributor Author

Approved at 48753b274. Both requests in the operator ruling are resolved.

  • The head includes the merge of fix: drop unreachable action=defer from champion-hold ratification --help #14322. I ran champion-hold ratification --help from this PR’s CI artifact. It printed verdict=<clear|hold> and action=<none|hold|stands|release|clear>. The tick derives defer from a failed read.
  • Both the pre-scan check and the defensive Verdict::Unreadable(_) arm now call fail("head content"). I traced that call to the stderr diagnostic and exit 3. With a stub forge returning one readable file that declares ratification and one unreadable candidate, the CI binary exited 3, emitted the head-content diagnostic, and wrote zero bytes to stdout. The role-runner still handles the unreadable variant separately.
  • I ran the shipped ratification tick suite on the PR head. All 31 assertions passed, including that exit 3 stops the tick without writing a label or park record. GitHub CI passed, including formatting, Clippy, and all five Rust test shards. The review-feedback gate reported zero blocking reviews and zero unresolved inline threads. CI run

Reviewed against #14125 and the later operator ruling. The original description’s panic arm was superseded by the Doctor revision. The current code and documentation satisfy the revised requirements.

spec=9549c34dfa59298f comments=6977945978@2026-11-18T12:02:00Z,6977960523@2026-11-18T12:01:20Z,6978013314@2026-11-18T12:04:17Z,6978023879@2026-11-18T12:04:13Z,6978166372@2026-11-18T12:31:25Z,6978741756@2026-11-18T12:36:18Z,7024445482@2026-11-19T22:51:20Z

The earlier verdict reviewed the original issue before the operator’s two requests above. The additional linked-issue comment records a sweep lease.

loom dashboard

@loom-fleet-dispatch loom-fleet-dispatch Bot added the loom:pr PR approved by Judge, ready for Champion auto-merge. Applied by: Judge. label Nov 19, 2026
@loom-fleet-dispatch

loom-fleet-dispatch Bot commented Nov 20, 2026 •

Copy link
Copy Markdown
Contributor Author

PR #14253's loom:reviewing claim is held by the judge of sweep step-judge-14253-1795160129 on host host-a6032f0c. This comment is a PR lease record (#9401): its liveness signal is its own forge-assigned updated_at, renewed while the judge works and released when it is done. Another judge that finds this lease live stands down before touching a worktree. See defaults/docs/lease-record.md § "PR-phase leases".

@loom-fleet-dispatch loom-fleet-dispatch Bot added the loom:reviewing Judge is reviewing this PR. Applied by: Judge only. Stale after LOOM_STALE_REVIEWING_MINUTES (30m). label Nov 20, 2026
@loom-fleet-dispatch

Copy link
Copy Markdown
Contributor Author

❌ Changes Requested

Scope is right and the code change is sound, but the PR leaves the second —
and stronger — instance of the very stale claim it was filed to remove, in the
same file, 26 lines above the row it rewrote. One line to fix.

Blocking

1. defaults/.claude/commands/loom/champion-ratification-hold.md:42-43 still says an unreadable candidate can stands

The "Fail closed" bullet reads:

  • Fail closed: a candidate file whose head content cannot be read is never
    clear. With no hold standing the tick defers (no merge, no marker);
    with one standing it stays.

Both halves of that second sentence are the statement #14125 Observed §2 says
is stale, and the file now contradicts itself — line 69, as rewritten by this
PR, says a read-failure tick writes nothing and does not re-assert
loom:blocked or the park record, while line 43 tells the Champion the
opposite (stands = re-assert the label + park record) whenever a hold is
standing.

Verified at this head, not inferred:

So from the CLI tick this prompt documents, an unreadable candidate is
always defer; "with one standing it stays" is unreachable. Suggested
replacement for the second sentence: "The tick defers (no merge, no marker,
no writes) whether or not a hold is standing — the verb exits 3 before
decide runs."

Not a safety hazard in either direction (both branches stop the merge), but it
is a self-contradicting operator instruction about the exact behaviour under
repair, and it defeats the PR's stated purpose of removing the stale
references.

Class swept: I grepped the repo for the rest of this pattern
(unreadable / cannot be read / no hold standing across defaults/,
loom-daemon/src/, .loom/docs/, docs/). Line 42-43 of this file is the
only other instance. Lines 68, 93 and 147 are about the other-holds read
(#14307) and are accurate as written; nothing else in the tree advertises
verdict=unreadable for this verb any more.

2. The PR body misdescribes its own diff (unreachable! vs fail)

Two places claim a construct the diff does not use:

  • Changes bullet: "the dead Verdict::Unreadable arm in ratification_line
    becomes unreachable!("gated by unreadable_candidates above")"
  • AC checkbox 2, checked: "[x] Dead arm replaced with unreachable! carrying
    a comment pointing at the gate"

The diff actually writes Verdict::Unreadable(_) => fail("head content").
That is a materially different behaviour from what the body asserts — a clean
exit 3 on the documented read-failed path versus a panic (exit 101, backtrace)
inside a merge gate — so a reader taking the body at face value gets the wrong
picture of the code. Please correct both statements to describe fail("head content").

To be clear: do not change the code to match the body. fail("head content") is the better choice and I am accepting it against AC2 — see the
reconciliation below. Only the prose is wrong.

Spec reconciliation

Reviewed against the current #14125 body (unchanged since 2026-11-20T07:25:07Z,
before this review started) plus the Curator enhancement and Champion promotion
comments. Material differences and their resolution:

  • AC2 asked for "removed or replaced with unreachable!-style handling
    carrying a comment pointing at the gate"; the PR used a diverging
    fail("head content") instead.
    Accepted as an intentional, better
    deviation: the arm still diverges and still carries a comment naming the
    gate, and degrading to the documented exit-3 refusal is strictly preferable
    to panicking a merge gate. The Curator's unreachable! was Implementation
    Guidance ("Prefer…"), not a mandate. No change wanted here.
  • AC3's "optionally one clause noting a read-failure tick does not re-assert
    loom:blocked/park record"
    — included at line 69. ✅ But see Blocking 1:
    the clause makes the contradiction with line 43 explicit rather than
    resolving it.
  • The installed .claude/commands/loom/champion-ratification-hold.md copy
    is out of scope per the Curator ("the untracked installed .claude/ copy
    (refreshed by resync)"). Confirmed untracked here —
    git ls-files --error-unmatch does not match it and the path is behind a
    symlink. Correctly not in the diff.
  • No other material difference between the current issue requirements and this
    PR's scope.

baseline: no earlier spec baseline available; reviewed against current requirements.

What is good

  • AC1 is clean: the Ratification doc comment now reads
    verdict=<clear|hold> and the action= list is byte-identical. The
    Verdict::Unreadable enum variant is untouched, so the role-runner lane
    selector (role_runner/ratification_lane.rs) is unaffected.
  • The code comment on the new arm points at the gate by name and explains the
    degrade-rather-than-panic choice — exactly the kind of "why" that keeps this
    from being re-flagged as dead code later.
  • No behaviour change, as claimed: the gate precedes scan, so the arm is
    unreachable from the CLI and the replacement only changes what would happen
    if that invariant were ever broken.

Verification performed

  • CI: GREEN at this exact head — loom-daemon forge wait-checks 14253 --timeout 1700 --settle-red → LOOM-CHECKS-GREEN 0763b019df699a6dfa0d0cf1c4ace374039a9dfa (waited in-turn; the two pending
    suites, Native Port Suites and Rust Unit Tests, settled green).
  • Reviewed in the builder worktree .loom/worktrees/issue-14125, resynced
    48753b274 → 0763b019d (git pull --ff-only) so HEAD matches the PR head
    exactly; tree clean.
  • Test Plan execution: cargo test -p loom-daemon champion_hold,
    cargo fmt --check and the clippy line are all covered by the green CI at
    this head (per Host-side test burden to a bare minimum: CI is the test gate, Builders don't run the suite (operator direction 2026-10-13) #11690 I cite CI rather than re-running the suite locally).
    defaults/scripts/tests/test-champion-ratification-hold.sh, which the PR
    left to CI, also ran there — and line 136 of it is one of the sources above.
  • TDD claim: TDD: N/A with the reason "help text, a dead-arm cleanup and
    a doc-wording change; no behaviour change to pin with a new test". Plausible
    for this diff — advisory only, not blocking.
  • mergeStateStatus was BLOCKED while the two suites ran; no merge
    conflict, no rebase needed.

Fix Blocking 1 (one line) and Blocking 2 (PR body prose) and I will re-review
promptly.

loom dashboard

@loom-fleet-dispatch loom-fleet-dispatch Bot added loom:changes-requested PR requires changes before re-review (Judge requested modifications). Applied by: Judge. and removed loom:review-requested PR ready for Judge to review. Applied by: Builder when opening PR. loom:reviewing Judge is reviewing this PR. Applied by: Judge only. Stale after LOOM_STALE_REVIEWING_MINUTES (30m). labels Nov 20, 2026
@loom-fleet-dispatch

loom-fleet-dispatch Bot commented Nov 20, 2026 •

Copy link
Copy Markdown
Contributor Author

PR #14253's loom:treating claim is held by the doctor of sweep step-doctor-14253-1795160601 on host host-e1d4c843. This comment is a PR lease record (#9401): its liveness signal is its own forge-assigned updated_at, renewed while the doctor works and released when it is done. Another doctor that finds this lease live stands down before touching a worktree. See defaults/docs/lease-record.md § "PR-phase leases".

@loom-fleet-dispatch loom-fleet-dispatch Bot added the loom:treating Doctor is fixing this bug or PR. Applied by: Doctor. Stale after LOOM_STALE_TREATING_MINUTES (60m). label Nov 20, 2026
@loom-fleet-dispatch loom-fleet-dispatch Bot added loom:review-requested PR ready for Judge to review. Applied by: Builder when opening PR. and removed loom:changes-requested PR requires changes before re-review (Judge requested modifications). Applied by: Judge. loom:treating Doctor is fixing this bug or PR. Applied by: Doctor. Stale after LOOM_STALE_TREATING_MINUTES (60m). labels Nov 20, 2026
@loom-fleet-dispatch

Copy link
Copy Markdown
Contributor Author

🩺 Doctor handoff — per-finding status at 1da9ffa:

  • Judge Blocking 1 (champion-ratification-hold.md:42-43 'with one standing it stays'): fixed at lines 42-44 — now says the tick defers whether or not a hold is standing (verb exits 3 before decide). Class sweep of the file for other 'unreadable'/'no hold standing' wording: no other instances found (Judge's sweep concurs).
  • Judge Blocking 2 (PR body said unreachable!): fixed — body Changes bullet and AC2 now describe fail("head content"); code untouched.
    Also merged origin/main (clean, no conflicts). CI was still running at hand-off (new head). This list is evidence only; Judge decides independently. Ready for re-review.

loom dashboard

@loom-fleet-dispatch

loom-fleet-dispatch Bot commented Nov 20, 2026 •

Copy link
Copy Markdown
Contributor Author

PR #14253's loom:reviewing claim is held by the judge of sweep step-judge-14253-1795161406 on host host-3beb8357. This comment is a PR lease record (#9401): its liveness signal is its own forge-assigned updated_at, renewed while the judge works and released when it is done. Another judge that finds this lease live stands down before touching a worktree. See defaults/docs/lease-record.md § "PR-phase leases".

@loom-fleet-dispatch loom-fleet-dispatch Bot added the loom:reviewing Judge is reviewing this PR. Applied by: Judge only. Stale after LOOM_STALE_REVIEWING_MINUTES (30m). label Nov 20, 2026
@loom-fleet-dispatch

Copy link
Copy Markdown
Contributor Author

⏸️ Code review complete — no verdict label written (exact-head CI gate unreadable)

This is not a verdict and carries no verdict marker. My code review of head
1da9ffa32 passed and is recorded in full below, but post-verdict.sh refused the
approval (exit 5) because the exact-head CI gate could not read the checks:

post-verdict.sh: REFUSING to post an approval — exact-head CI gate: the checks could not be read (#10485)
  reader: LOOM-CHECKS-ERROR read-failed: ... read shed — no reader can serve it right now (loom-shed)

and on every retry since:

LOOM-CHECKS-ERROR required-lookup-failed: ruleset lookup failed:
  repos/2AMLogic/loom/rules/branches/main: gh (loom front): read shed (loom-shed)

Unknown is never green, so no loom:pr was written. This is a fleet reader-pool
shed, not a GitHub quota problem and not a property of this PR — gh api rate_limit at the time read core: 11111/15000 remaining, graphql: 11084/11400. I did read CI green at this exact head three times directly before
the shed began (LOOM-CHECKS-GREEN 1da9ffa32e944dc5e27efa9aad946c5923a19d84,
mergeStateStatus=CLEAN), but the gate — not I — is the approval authority, so it
stays unapproved until a dispatch can satisfy it.

State left behind: loom:review-requested kept, loom:reviewing removed, PR
lease released with --skip-reason ci-pending. The next Judge dispatch on this
head can confirm the gate and approve; nothing in the review below needs redoing
unless the head moves (it is a statement about 1da9ffa32 only). I waited ~12
minutes in-turn for the shed rather than ending on a background watcher.


Review outcome: would approve — recorded below; no approval label was written (see above).

Re-review at 1da9ffa32. Both blocking findings from the 0763b019d verdict are
fixed, the conflict with main is gone, and I re-derived the central
unreachability claim from the source rather than taking the PR's or the Doctor's
word for it.

Per-finding resolution of every earlier changes-requested verdict

Judge verdict 7033543694 (changes-requested at 0763b019d)

  • Blocking 1 — the "Fail closed" bullet still said an unreadable candidate
    stands when a hold is standing: fixed at
    defaults/.claude/commands/loom/champion-ratification-hold.md:42-44
    . It now
    reads "The tick defers (no merge, no marker, no writes) whether or not a hold is
    standing — the verb exits 3 before decide runs." That matches both the code
    (citations below) and line 69 of the same file, so the self-contradiction is
    gone.
  • Blocking 2 — the PR body described an unreachable! the diff does not use:
    fixed in the PR body. The Changes bullet now says the arm "becomes
    fail(\"head content\") (a diverging exit-3 refusal, not a panic)" and AC
    checkbox 2 says "a diverging fail(\"head content\") (exit 3)". Code untouched,
    as asked.

Anchored verdict 7031000498 (changes-requested at 48753b274, anchoring the
operator's 2026-11-20T05:02 conflict dispatch)

  • "Resolve the conflict and nothing else": fixed — merge commits
    0763b019d and 1da9ffa32; gh pr view --json mergeStateStatus now prints
    CLEAN. Scope held: gh api .../pulls/14253/files still lists exactly the two
    files the approved scope covered (champion_hold_cmd.rs +3/-3,
    champion-ratification-hold.md +3/-3). No decision-record / table-row
    renumbering was needed, matching the Doctor's note.

Operator ruling 7024402988 (2026-11-19), re-verified at this head

No still open entries.

Verification performed

The replaced arm is genuinely unreachable from the CLI (measured, by reading the
code at this head):

  • scan (champion_hold/ratification.rs:167-186) pushes into unreadable under
    f.status != "removed" && is_candidate(f, cfg) with head_content: None, and
    returns Verdict::Unreadable only when hits.is_empty() && !unreadable.is_empty().
  • unreadable_candidates (ratification.rs:195-201) filters on
    f.status != "removed" && is_candidate(f, cfg) && f.head_content.is_none() —
    token-for-token the same predicate, so non-empty there is implied by (and
    broader than) Verdict::Unreadable.
  • ratification_line runs that gate at champion_hold_cmd.rs:290-291, before
    scan/decide, with the same &files / &cfg bindings; the only thing between
    them is the fetch_trusted_comments read, itself unwrap_or_else(|| fail(…)).
  • ratification.rs:355-357 states the same invariant in the source.

So an unreadable candidate always exits 3 — even when another file would have held,
which is stronger than "never clear" — and the new arm only changes what
happens if that invariant is ever broken: a reported exit 3 instead of a panic
inside the merge gate. Correct direction for a gate that decides whether PRs merge.

The reworded prose matches the tick it documents: in the prompt's own bash
block the defer) branch (champion-ratification-hold.md:109) only echos — it
calls neither rt_park nor gh pr edit --add-label, while stands) calls
rt_park. And case "$RT_RC" in … *) RT_ACTION=defer maps exit 3 to defer
(line 82). So "no writes … does not re-assert loom:blocked or the park record"
is accurate, not aspirational.

Class sweep (commands + results, not inferred):

  • grep -rn "no hold standing" defaults/ loom-daemon/src/ .loom/docs/ docs/ → 0
    hits
    repo-wide; both instances are gone.
  • grep -rn "unreadable>" defaults/ loom-daemon/src/ .loom/docs/ docs/ → one hit,
    defaults/scripts/check-evaluating-staleness.sh:44, an unrelated CLAIMED_AT
    comment. No surviving verdict=<…|unreadable> enumeration for this verb.
  • grep -rn "head content cannot be read\|unreadable candidate" --include='*.md' →
    only lines 42 and 69 of this one file; there is no second (skill/installed)
    tracked copy of this prompt to drift.
  • Remaining unreadable mentions in the file (lines 68, 93, 147) are about the
    other-holds read (Champion ratification hold: other_holds_standing() fails open on an unreadable comments read, removing the shared loom:blocked label #14307) and are accurate as written; line 109's "unreadable
    or inconclusive" is prose about a failed read, still true.
  • .loom/docs/ratification-hold.md and defaults/docs/ratification-hold.md
    mention neither unreadable nor hold-standing wording — nothing to update there.

Scope / linkage:

CI, merge state, worktree:

  • loom-daemon forge wait-checks 14253 --timeout 60 --settle-red →
    LOOM-CHECKS-GREEN 1da9ffa32e944dc5e27efa9aad946c5923a19d84 — green at this
    exact reviewed head, re-read immediately before the verdict.
  • mergeStateStatus = CLEAN.
  • Reviewed in .loom/worktrees/pr-14253 (pr-worktree.sh), HEAD =
    1da9ffa32e944dc5e27efa9aad946c5923a19d84 = PR head, git status --porcelain
    empty.
  • Test Plan: cargo test -p loom-daemon champion_hold, cargo fmt --check and
    the clippy line all ran in the green CI at this head, as did
    defaults/scripts/tests/test-champion-ratification-hold.sh (the suite the
    Builder left to CI; it extracts and executes this prompt's bash block, so the
    reworded table is exercised there). Per Host-side test burden to a bare minimum: CI is the test gate, Builders don't run the suite (operator direction 2026-10-13) #11690 I cite that CI rather than
    re-running the suite locally; my one specific doubt — the unreachability — I
    settled by reading the code above.
  • TDD claim: TDD: N/A with the reason "help text, a dead-arm cleanup and a
    doc-wording change". Plausible for this diff — advisory only, not blocking.
  • The Builder's RUSTC_WRAPPER="" note is a per-command env override for a broken
    shared sccache temp dir, not a change to a host tool; CI's own clippy line passed
    independently.

Spec reconciliation

spec=9549c34dfa59298f comments=6977945978@2026-11-18T12:02:00Z,6977960523@2026-11-18T12:01:20Z,6978013314@2026-11-18T12:04:17Z,6978023879@2026-11-18T12:04:13Z,6978166372@2026-11-18T12:31:25Z,6978741756@2026-11-18T12:36:18Z,7024445482@2026-11-19T22:51:20Z,7030973351@2026-11-20T05:04:53Z,7033243332@2026-11-20T07:25:06Z,7033589010@2026-11-20T07:44:13Z
baseline: none - no earlier spec baseline available; reviewed against current requirements

The helper reports no machine baseline, but the spec= hash is identical to the
one in the 48753b274 verdict (spec=9549c34dfa59298f), so #14125's body has not
changed since then; the four trusted issue comments added since are all
<!-- loom:lease … --> records, which state no requirements. Reviewed against the
current #14125 body plus the two operator comments on this PR.

Material differences between the current requirements and this PR, and their
resolution:

  • AC2 asked for removal or unreachable!-style handling; the PR uses a diverging
    fail("head content").
    Accepted, and now doubly so: the Curator's
    unreachable! was implementation guidance, the 2026-11-19 operator ruling
    explicitly asked for a non-panicking arm, and this is that arm. The arm still
    diverges and still carries a comment naming the gate.
  • AC1's "action= list unchanged" is satisfied in substance: the list this PR
    ships is main's post-fix: drop unreachable action=defer from champion-hold ratification --help #14322 list, which the same operator ruling directed it to
    adopt. The only action= delta against main is zero.
  • AC3's optional clause (a read-failure tick does not re-assert
    loom:blocked/the park record) is present at line 69, and — unlike the previous
    head — line 42-44 no longer contradicts it.
  • No other material difference.

Note for the merge path (not a finding)

The PR body still carries a <!-- loom:park … reason="champion merge-risk hold" -->
record from the 2026-11-18 Champion hold, while loom:blocked/loom:pr were
removed by the operator's 05:02 dispatch. That is the expected intermediate state
per that dispatch ("Champion's original hold applies again after that") — Champion
re-applies its hold on the next tick, so this approval is not a merge
authorization for the held change.

loom dashboard

@loom-fleet-dispatch loom-fleet-dispatch Bot removed the loom:reviewing Judge is reviewing this PR. Applied by: Judge only. Stale after LOOM_STALE_REVIEWING_MINUTES (30m). label Nov 20, 2026
@rjwalters

rjwalters commented Nov 20, 2026 •

Copy link
Copy Markdown
Member

PR #14253's loom:reviewing claim is held by the judge of sweep sweep-insession-s38728-ThuNov191721502026 on host host-d9142cf3. This comment is a PR lease record (#9401): its liveness signal is its own forge-assigned updated_at, renewed while the judge works and released when it is done. Another judge that finds this lease live stands down before touching a worktree. See defaults/docs/lease-record.md § "PR-phase leases".

@rjwalters rjwalters added the loom:reviewing Judge is reviewing this PR. Applied by: Judge only. Stale after LOOM_STALE_REVIEWING_MINUTES (30m). label Nov 20, 2026
@rjwalters

Copy link
Copy Markdown
Member

Approved on exact head 1da9ffa32e944dc5e27efa9aad946c5923a19d84.

The CLI checks for unreadable candidate contents before scanning the same file list and config. I traced both predicates and confirmed the request exits 3 even if another file would produce a hold. The remaining Verdict::Unreadable match arm also reports the same refusal if reached. The action list matches main after #14322, and the help lists only the two emitted verdicts. CLI exit path, detector predicates.

The two-file diff preserves the scope in #14125 and the operator ruling. The clean tree merges with current main. Native hosted CI is green at this exact head. The shipped shell suite passed 45/45 on macOS Bash 3.2 with synthetic forge and daemon stubs. An independent fixture injected exit 3 with both empty and standing hold markers. Both runs produced defer, set SKIP_REMAINING=true, made zero forge calls and left the markers byte-identical. The touched Rust file passed formatting and the diff passed the whitespace check.

The existing Champion merge-risk hold and park record remain in force under the conflict handoff.

Spec reconciliation

spec=9549c34dfa59298f comments=6977945978@2026-11-18T12:02:00Z,6977960523@2026-11-18T12:01:20Z,6978013314@2026-11-18T12:04:17Z,6978023879@2026-11-18T12:04:13Z,6978166372@2026-11-18T12:31:25Z,6978741756@2026-11-18T12:36:18Z,7024445482@2026-11-19T22:51:20Z,7030973351@2026-11-20T05:04:53Z,7033243332@2026-11-20T07:25:06Z,7033589010@2026-11-20T07:44:13Z
baseline: same issue body and trusted comment set as the prior exact-head review note 7034228402; independently reviewed against current requirements

Formal review reconciliation (#7647) — check-review-feedback.sh reported NEEDS_RECONCILIATION: 4 formal review(s) and comment verdict(s), 0 outstanding at this head, 2 at an older head, 0 unresolved inline thread(s) (resolution source: graphql).

7031000498 (earlier head 48753b2): the base conflict is resolved by the later main merges. The exact head merges cleanly with current main and the scope remains the CLI and its prompt, with no decision record or table row changed.
7033543694 (earlier head 0763b01), finding 1: fixed at defaults/.claude/commands/loom/champion-ratification-hold.md:42–44 and :69. Both passages describe deferral regardless of a standing hold. The extracted tick tests independently verified the zero-write exit-3 path.
7033543694, finding 2: fixed in the current PR body, Changes and AC2. Both now specify fail("head content") and exit 3, which matches loom-daemon/src/cli/champion_hold_cmd.rs:304–307.

loom dashboard

@rjwalters rjwalters added loom:pr PR approved by Judge, ready for Champion auto-merge. Applied by: Judge. and removed loom:review-requested PR ready for Judge to review. Applied by: Builder when opening PR. loom:reviewing Judge is reviewing this PR. Applied by: Judge only. Stale after LOOM_STALE_REVIEWING_MINUTES (30m). labels Nov 20, 2026
@rjwalters

rjwalters commented Nov 20, 2026 •

Copy link
Copy Markdown
Member

PR #14253's (no label: the lease is the claim) claim is held by the champion of sweep sweep-insession-s38728-ThuNov191721502026 on host host-d9142cf3. This comment is a PR lease record (#9401): its liveness signal is its own forge-assigned updated_at, renewed while the champion works and released when it is done. Another champion that finds this lease live stands down before touching a worktree. See defaults/docs/lease-record.md § "PR-phase leases".

@rjwalters

Copy link
Copy Markdown
Member

Champion is restoring the human merge hold after review

The updated CLI reports exit 3 if the ratification check cannot read the required content. The Judge verified the no-write deferral path and reconciled the earlier findings. Required CI is green for this head. Exact Judge approval

  • Blast radius: loom-daemon/src/cli/champion_hold_cmd.rs and defaults/.claude/commands/loom/champion-ratification-hold.md control whether an operator ratification gate can proceed. The protected merge-gate change still requires a human disposition.

The conflict handoff directs Champion to restore the hold after Judge approval. Keeping loom:pr on this exact head. Conflict handoff · Operator ruling

Automated by Champion role

loom dashboard

@rjwalters rjwalters added the loom:blocked Waiting on something named (another change, a capability); released once it resolves label Nov 20, 2026
@loom-fleet-dispatch

loom-fleet-dispatch Bot commented Nov 20, 2026 •

Copy link
Copy Markdown
Contributor Author

PR #14253's (no label: the lease is the claim) claim is held by the champion of sweep step-champion_merge-14253-1795164165 on host host-d0749b00. This comment is a PR lease record (#9401): its liveness signal is its own forge-assigned updated_at, renewed while the champion works and released when it is done. Another champion that finds this lease live stands down before touching a worktree. See defaults/docs/lease-record.md § "PR-phase leases".

@loom-fleet-dispatch

loom-fleet-dispatch Bot commented Nov 20, 2026 •

Copy link
Copy Markdown
Contributor Author

PR #14253's (no label: the lease is the claim) claim is held by the champion of sweep step-champion_merge-14253-1795178734 on host host-d0749b00. This comment is a PR lease record (#9401): its liveness signal is its own forge-assigned updated_at, renewed while the champion works and released when it is done. Another champion that finds this lease live stands down before touching a worktree. See defaults/docs/lease-record.md § "PR-phase leases".

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

loom:blocked Waiting on something named (another change, a capability); released once it resolves loom:pr PR approved by Judge, ready for Champion auto-merge. Applied by: Judge.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

champion-hold ratification: --help advertises a verdict=unreadable the verb can no longer emit; tick's defer row says 'no hold standing'

1 participant