stack/agents/filbert/work/queue-156/r2/body.md at dev-console

0 Fork 1

Files

T

jason.woltje b9ea9304fd lead: commit filbert's untracked evidence (273 files: queue-101 queue-102 queue-105 queue-107 queue-108 queue-109 queue-110 queue-112 queue-113 queue-114 queue-115 queue-116 queue-119 queue-120 queue-122 queue-123 queue-125 queue-126 queue-128 queue-129 queue-130 queue-131 queue-132 queue-133 queue-134 queue-135 queue-136 queue-137 queue-139 queue-140 queue-141 queue-142 queue-145 queue-146 queue-147 queue-148 queue-149 queue-150 queue-151 queue-152 queue-153 queue-154 queue-155 queue-156 queue-162 queue-163 queue-165 queue-166 queue-167 queue-169 queue-170 queue-175 queue-177 queue-181 queue-182 queue-184 queue-185 queue-186 queue-187 queue-188 queue-49 queue-64 queue-66 queue-70 queue-71 queue-80 queue-84 queue-85 queue-87 queue-88)

Unclaimed files left in the canonical checkout. Inventory: canonical-cleanup-inventory.md (commit-as-evidence groups). detect-secrets scan clean on 2026-11-02: hits are sha256 pins, git object ids, test sentinels already in tracked tests and unit-file lines.

2026-11-02 15:05:14 -06:00

6.0 KiB

Raw Permalink Blame History

Filbert review, row 156 round 2: approve

Candidate manifest build-manifest-r2.sha256, sha256 72a5c1b62ed0cb7be43015f23aca835ccec001081ff332d8d8c4fd28bd2bb700 (14 paths, the same as round 1). Patch build-r2.patch, sha256 a9fa9b75c3f7c0d91ad2b410f681c7e5b8258ecfc07cc646f3393ccfe5d98e48. I applied it with --index to 70c9131f0. All 14 manifest lines matched, git diff --cached --check is clean, and the diff is +925 -92 over 14 files.

What changed since round 1

I built the round 1 tree (build.patch on d88613bac) and compared it with this one file by file. Eight files differ:

broker.mjs, launcher.mjs (one comment), cli/README.md;
the tests land.test.mjs, launcher.test.mjs, thread-launch.test.mjs and runner.test.mjs;
docs/TOOLS.md, where the only difference is row 154's landed :443 hunk from the base (d88613bac..70c9131f0), not this row.

The other six files are byte-identical to round 1, so my round 1 "checked and fine" items carry over.

F1: resolved

broker.mjs:180 adds #received, keyed business:run.
message.receive adds the run after #agent(s) passes (:1844-1846), so humans and testers are never added.
landRelease answers {released:false, landing:true} for a run not in the set (:1016). That check comes after the run, holder and decision checks and before the unread count, so a refusal still names its code.
The runner takes every landing turn (runner.mjs:378-400) before its first message.receive (:404), and nothing between role.claim and that loop calls message.receive. So a run in a landing turn is never released.
The cli/README.md text now says this. Its "only between turns" sentence and TOOLS.md's "stopped between turns first" now hold as written.

The test is the one I asked for. The runner.test "a run in a landing turn isn't released..." test uses a real runner and broker:

the adapter holds landing turn 1 open on a file;
d2 is raised and resolved during that turn, and landHolders lists both decisions;
two landRelease calls 300 ms apart answer landing: true, and the claim stays;
after the go file, it releases by land-relaunch only once both DECISION messages are read, and the runner exits not-holder.

launcher.test :489-497 shows that a tick given landing: true stops nothing and logs no land-relaunch. Dewey's "tick mapping" splits that coverage honestly between the two tests. I accept the split: the runner test calls landRelease with the arguments the tick uses.

F2: resolved

thread-launch.test :306-321 answers landHolders with an entry for the thread instance, launches a thread run, and asserts session.json has no landing key. The BUILD.md:56 equivalence claim is struck and corrected under "Equivalent mutants".

F3: resolved

land.test :454-475 ends the raising run, binds and claims a new run (merge-gate-land), and checks two things before any message.receive:

landRelease answers landing: true;
checkLand returns a ticket naming the decision, commit, verdict and choice, and recordLand records the merge with actor_run merge-gate-land.

Afterwards landHolders is empty, and no op='delivered' row exists for the run. Dewey's run 4 correction says why the assert counts delivered and not every row: the claim writes routed rows. That matches the release test's filter at :445. The call is made directly as the run, not through a runner's land_work, which sage-0942's F3 ruling allows.

Each new test kills its mutant

I read the mutant logs in run4/mutants/. Each failure is the new assert, not a crash:

B13 (drop the #received check). Fails 3 tests. The land.test release test and the F3 test fail at their landing: true asserts. The runner test fails at runner.test.mjs:739: released: true while the landing turn is open, which is round 1's defect.
B14 (drop the add on receive). Fails 2 tests. The land.test release test fails, and the runner test times out at :746 waiting for a release that never comes.
LA14 (a thread launch asks landHolders). The thread-launch test fails at :320, and the failure prints the session file with "landing":[{"decision":"d9",...}].

run4.out: killed 36, survived 0, timeout 0, mutants.sh rc=0, and the manifest matched at the end. My awk sums over Dewey's logs agree:

run4/touched (4 logs): tests 84, pass 84, fail 0, skipped 0;
run4/singles (26 logs): tests 290, pass 290, fail 0, skipped 0.

thread-launch's log shows skipped 0, so its { skip } option didn't skip the F2 test.

Other checks

Hash pins. Only thread-migration.test.mjs:46 has a 64-hex pin in a file that names a non-test manifest path, and it pins a DDL string, not broker.mjs.
Base drift. From 70c9131f0 to HEAD c3cd27345, no manifest path changed, and build-r2.patch passes git apply --cached --check on HEAD in a temporary index.

Notes (not blocking)

A run that never reaches its first message.receive (a hung landing turn) keeps the release closed. So does a run whose broker restarted, until its next receive, because #received is in memory. BUILD.md records sage-0944's fallback for the first case: row 149's path, where the chief ends the run and launches the holder. The second case clears at the run's next poll. Neither is in cli/README.md yet. Doc-only, follow-up.
#received is never pruned. It grows by one entry per role run that receives, the same as #ended. That's fine at this scale.

Ran by me

Nothing that needed the slot. I applied both rounds' patches in two worktrees, ran sha256sum -c (14/14), compared the files, ran the temporary-index apply check on HEAD, awk-summed Dewey's logs and read the mutant logs, and ran greps.

Not mine

Dewey's run 4 rerun on Sage's slot (22:53:04Z to 23:08:35Z, sage-0947): touched 84/84/0, singles 290/290/0, and 36 of 36 mutants killed. I read and summed the logs but did not rerun them. The red run 4 (land.test 1 fail) is in run4-red/, as Dewey recorded it. The full gate on the new base is Sage's at landing.

6.0 KiB Raw Permalink Blame History