Every push to dev paid for the full browser trio and the backend suite,
including commits that touch only documentation, workflows or process
scripts — the bundle and the harness were byte-identical, so the runs
proved nothing new. On 2026-08-19 alone that was roughly six pushes at
about seven minutes each.
The reuse key per heavy job is sourceFingerprint (src, demo fixtures,
golden scenarios, build manifests) plus that job's own harness: smoke
takes demo/smoke_*.mjs, golden takes demo/golden/** including baselines,
performance_smoke takes demo/performance/**, backend takes tests_backend
and the Python sources. A cache marker is written only by a successful run
of the same key, so a hit proves a job with identical inputs already
passed. scripts/** is deliberately outside every key: infrastructure work
edits it constantly and reuse would never fire.
This is not the path filter from the `changes` job, which stays disabled
on dev on purpose: there the scope is guessed from paths and "green" means
different things, here input equivalence is proven by a hash. And a
release candidate always bumps the version, which is part of the
fingerprint, so its keys are new by construction and the full gate set
still runs before every beta and release.
A waived job is announced with a notice and a run summary line rather than
skipped in silence, and the marker save tolerates a concurrent identical
run instead of reddening the job.
Issue: #208
User-Visible: no
Rule 8 demanded a working S-label from every class A/B commit's issue,
while owner decision #118 sends infrastructure work outside the S1..S8
flow entirely — such an issue has no status label by construction. The
two rules contradicted each other and the machine-checked one won, so
Validate on dev went red on every infrastructure commit (#175, #191,
#202, #206) and the catch-up signal stopped meaning anything. A gate that
is always red is not a gate.
The waiver keys on the diff, not on a permission label: a range with no
class A file at all. An `infra` label could be pinned on a product task
to walk a product commit past the status check; ceasing to touch class A
without ceasing to be infrastructure work is not possible. Issue
existence, open state, `blocked` and fail-closed on an unreachable gh all
still apply, and the waiver prints a visible warning rather than passing
in silence.
Mutation-checked both ways: unwiring the waiver reddens the CLI test,
and letting class A keep the waiver reddens both new tests.
Issue: #207
User-Visible: no
After the mandatory rebase of a published issue branch the pre-push hook
still passes remote_old..local_new, and once the old tip is no longer an
ancestor that range drags in the whole advanced dev history: on #117 it
meant 84 foreign commits and 20 false rule-8 rejections over already
closed issues, leaving --no-verify as the only exit.
The clamp lives in the gate rather than the hook: .githooks/pre-push
carries an executable bit that MCP publication strips (the commit-msg
precedent), so editing it needs an owner-side commit. When the target is
an issue branch and the declared base is not an ancestor of the head, the
base becomes the merge-base with origin/dev. Fast-forward pushes keep
their exact range, every own commit is still judged, and a real violation
in a post-rebase commit still blocks — covered by a scenario test that
goes red without the wiring.
Issue: #190
User-Visible: no
First real run of the gate failed before reaching a single mutant: the clean
run of the golden guard was red on untouched code. demo/golden/policy.mjs
refuses `verify --scenario=...` on purpose — a partial verify is the "make CI
green" loophole the policy exists to close. The gate built to catch dishonest
tests had reached for a dishonest shortcut, and the policy caught it.
capture keeps the whole check: a failed semantic assertion becomes status
error, and goldenRunFailed treats an error as failure in either mode. The scene
carries warmPixelRegion with minPixels 2500 over the receiving half, so a lamp
moved out of reach still fails it — which is exactly what this mutant asserts.
Issue: #85
User-Visible: no
The owner stopped using GitHub Projects. Most of this is wording, but one part
was not: release-prerelease.mjs talked to the Project in code. finishIssues
looked up the project id, listed its items and its Status=Done option, and threw
when an issue was missing from the board — so the first release that closed an
issue would have died on a step with nothing to do with publishing. Found by
reading rather than by releasing, which was luck.
Closing issues stays, and now strips the status label first. That order is not
cosmetic: the invariant that a closed issue carries no status label has broken
twice already, both times because a manual step did it the other way round. The
close-merged job already does it in this order.
The documents now say labels and only labels. The explicit "no longer used"
lines are kept on purpose, in PROCESS.md and next to the code that used to sync:
a decision that vanishes quietly gets reintroduced a month later by someone who
never knew it was made.
Issue: #139
User-Visible: no
Five times in this project a green test meant nothing was checked. The
continuity smoke stayed green after the entire mechanism it guards was cut out.
The golden scene created to protect doorway light was empty — 1,177 warm pixels
against 107,119, all of them icons. The shadow smoke passed while no shadow was
drawn. Each time the test had been written alongside the code, went green at
once, and nobody ever asked whether it could go red.
The gate makes that question routine. Each mutant is a few lines of patch that
reproduce a known breakage, plus the name of the test that must fail on it. A
worktree is patched, the bundle rebuilt, the guard run — and a guard that stays
green fails the gate. Six mutants cover the holes documented in #85; the anchors
are exact strings from today's source, so the registry cannot silently drift —
a unit test that runs with the ordinary suite refuses a stale anchor.
The full run rebuilds the bundle per mutant, so it lives in its own workflow,
before a stable release and on a weekly schedule, not in Validate. The rules for
new tests are written at the top of docs/TESTING.md, and the sixth of them is
the cheapest: an assertion that reads back the property the code just set is
not written at all.
Issue: #85
User-Visible: no
The owner's report: the process works but every stage takes a long time even on
simple bugs. Two causes, and neither was the one that first comes to mind.
The reviewer ran everything regardless. On #89 it installed Chromium, ran all 127
smoke files and a full golden capture — right for a task rated 10/10 for
complexity, absurd for a bug about a room divider. Full suites are the pre-beta
gate; the review now runs typecheck, unit and build always, and smokes, golden,
pytest or performance only where the diff and the AC call for them. The price of
narrowing it is honesty: the reviewer must list which gates it ran, which it did
not, and why, so a skipped gate is a visible decision rather than a silent one.
The reviewer also built its own environment out of model turns, with no npm cache
and no browser cache, paid for from the same forty-five minutes. The workflow now
installs dependencies and Chromium as ordinary cached steps, after switching to
the task branch so the lockfile is the branch's own.
Second, ceremony did not scale down. The light track makes a spec cheap; the new
trivial track does without one — S2-analysis straight to S5-ready, no spec review,
AC in the issue body. It is deliberately hard to qualify for: a bug on one surface,
no new UX contract, no migration, no i18n, no perf or touch effect, three checkable
AC at most, and expected behaviour already on record. Nothing left to decide is the
criterion that holds the whole thing up, and it cannot be met by feeling sure.
Code review is never skipped on either track. It is what stands in for testing
here, so it is the one stage speed may not buy.
Issue: #127
Issue: #128
User-Visible: no
Section 10.1 promised pre-push as the blocking gate that replaces pull requests.
The hook did not exist, so the document promised a check that was not there —
worse than saying nothing, because a promise like that gets relied on. Until now
process-gate ran only as the catch-up job in CI, which reports after the code is
already in dev.
The hook skips branch deletions and tags, and for a branch the remote has not
seen it measures from the merge-base with dev rather than from the root, or every
violation committed before the gate existed would make it impossible to pass. A
missing script does not block a push: old checkouts and worktrees have to stay
usable.
gh is optional on purpose. Reading issue status needs the network, and a hook
that cannot work on a train is a hook people switch off; offline it runs what it
can and CI does the strict pass.
The executable bit is the quiet part. Git skips a hook without +x and says
nothing — the gate reports success by being absent. Measured on a real push:
mode 644 produces zero lines from the gate and the push goes through, 755 stops
it. The API cannot set the bit, so install-hooks restores it on every install.
Issue: #121
User-Visible: no
Check 2 compared the Issue trailers against whatever branch the working tree
happened to be on, over whatever range it was given. Those two are not the same
set. After a rebase the CI range widens — `before` points at a discarded commit,
the merge-base slides back, and commits that belong to dev arrive carrying other
issue numbers. Every one of them then looks like a violation.
Running the gate over real history from issue/89 with a dev range produced 26
false refusals out of 26 commits, which would have reddened Validate on the next
force-push of any task branch.
The rule now reads origin/dev..HEAD for its own verdict and leaves the event
range to the other checks. A commit that genuinely carries the wrong trailer for
its branch is still caught; the integration test covers both directions.
Issue: #105
User-Visible: no
PROCESS.md 10.2 describes scripts/process-gate.mjs; the script never existed.
Commits go straight to dev without PRs and GitHub blocks nothing on its side, so
until now the only thing standing between the process and rule #1 was the good
faith of whoever was committing. Hooks catch a violation on the author's machine
but --no-verify walks past them; this job is the catch-up pass that cannot be
skipped locally.
Checks 1-7 offline, 8 through gh, plus the escalation of check 3: a class A
commit with neither a spec file nor the `small` label is a failure, not a
warning. Check 8 is fail closed — an unreachable or closed issue is a refusal,
never a silent pass.
Two things surfaced while wiring it up and are recorded in the script header.
S8-merged had to join the allowed statuses: the pipeline merges into dev before
it moves the label, so Validate reads the issue already advanced and a strict set
would redden every accepted task. And the status question now applies only to
class A/B commits — asking it of a review document would fail every time, since
that document lands while the issue sits in S4-spec-review or S7-code-review.
Issue: #105
User-Visible: no
Sergey Matyunin