Skip to content

ci: feed the swarm from CI, not from whichever laptop is awake - #4281

Merged
gHashTag merged 1 commit into
masterfrom
ci/feed-empty-bodies
Sep 19, 2026
Merged

gHashTag merged 1 commit into
masterfrom
ci/feed-empty-bodies

Conversation

@gHashTag

Copy link
Copy Markdown
Owner

Closes #4280

The empty-bodies feeder moves off the operator's machine, where it fired twice in twelve hours on 2026-09-17 and zero times since 07:39 that day. Same measurements, same refusal to quote a number it has not executed; the compiler is the one the job builds.

Verified on this checkout with a real t27c: 29 files hold 154 empty bodies; 127 files covered by open issues; 0 uncovered.

🤖 Generated with Claude Code

The empty-bodies feeder fired twice between 18:03 and 05:43 on 2026-09-17
instead of 36 times (the machine slept), and zero times since 07:39 that day:
every run failed in three seconds on a model the provider does not serve, and
a script that never starts writes no log.

The script moves into the repository and runs on a schedule. What it measures
with is the compiler the job builds, because the same script pointed at an
eight-day-old binary built from uncommitted source counted 483 empty bodies in
139 files where master's compiler counts 245 in 70. What it claims it checks:
every command an issue quotes is executed first, as the exact string a bee will
read, and the issue is not opened unless each prints the claimed value.

Closes #4280

Gates: python3 -m py_compile on the ported script; run against this checkout
with a real t27c it reports "29 files hold 154 empty bodies; 127 files covered
by open issues; 0 uncovered" and creates nothing, matching the laptop version
on the same commit; python3 scripts/ci/check_pr_branch_filters.py exits 0 with
the new workflow classified.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@gHashTag
gHashTag enabled auto-merge (squash) September 19, 2026 15:18
@github-actions

Copy link
Copy Markdown
Contributor

📓 NotebookLM Notebook linked to this PR

This notebook contains session context, decisions, and artifacts for this work.

@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-09-19 15:18:53 UTC

Summary

Status Count
Total Open PRs 39
PRs with Failing Checks 18
PRs with All Checks Green 21
READY 20
FAILING 18
PENDING 0
NO CHECKS YET 0

These columns do not partition: 20 + 18 + 0 + 0 = 38, and there are 39 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=403499176a5d != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

@gHashTag
gHashTag merged commit f566a48 into master Sep 19, 2026
26 of 29 checks passed
gHashTag added a commit that referenced this pull request Sep 20, 2026
… path (#4304)

Three checks are red on master. Two are this loop's.

CENSUS. tools/census/quiet.txt pinned `workflow files read 51`; the tree holds
57. The six added since the pin are #4277, #4279, #4281, #4283, #4291 and
#4300, all merged by this loop, and cli-tri has been red since 2026-09-17.
What moved, explicitly: quiet 51 -> 57 workflow files and 123 -> 135 "named a
path but not quiet"; shell 51 -> 57 files, 72 -> 78 jobs, 249 -> 267 run-steps,
228 -> 246 steps the runner runs with `bash -eo pipefail`. The gates did not
change; the population grew.

HOME PATH. feed_defects.py carries a comment explaining that an absolute
`/Users/<name>/t27/target/release/t27c` once went into 137 open issues -- and
spells it out to say so, which is what the scan catches. It now reads
`.../t27/target/release/t27c`, same meaning, no home directory.

WORKTREES. That gate also failed in every worktree on a path no commit
contains: it excludes the `.git` DIRECTORY, and a worktree's `.git` is a FILE
holding `gitdir: /Users/...`. `--exclude=.git` joins `--exclude-dir=.git`. The
reviewer measures in a detached worktree and every bee works in one.

NOT FIXED HERE: `coverage` reports 544 stale seals and has no successful run in
its last 60 on master, reaching back past 2026-09-18. Re-sealing 544 specs
records what the compiler now produces -- a decision about the ledger, not a
repair.

Closes #4303

Co-authored-by: Claude <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

The empty-bodies feeder runs on a laptop, and has fired zero times since 2026-09-17 07:39

1 participant