No Full Performance profile walked the backdrop (imagePlan) path: every
large-house fixture has plan_url null. So the #739 K1 double render -- a
warm 2.5D floor switch with a backdrop cleared the ready paper, inserted
the veil, probed the card background and rendered a second time -- was
invisible to CI by time and structurally; a temporary probe found it.
large-house-isometric-backdrop-v1 is the twin of large-house-isometric-v1
with the shipped f1.svg under a URL of its own on every floor
(plan_aspect 1, room geometry unchanged). The variant is derived in the
runner as plan-snap's is, so demo/fixtures and the bundle fingerprint do
not change. A first stable frame without the backdrop image fails the
sample. After the switchCycle window and its #735 guard, before forced
GC and outside every timed window, a probe makes six warm switches and
counts performUpdate passes until updateComplete resolves true: the K1
second pass starts after the first updateComplete resolves, so a count
taken right after the first await reads one on both sides.
evaluate.mjs rejects a candidate of this profile unless every switch took
one pass, or when the probe is missing; the base is reported, not judged
(v1.78.0 and dev before #739 take two). The budget is a copy of the
historical isometric budget under the new profile id. performance.yml
gains the isometric-backdrop matrix entry with exact-SHA comparison.
Witness: a tree with #739 reverted reads perSwitch 2 in every sample and
benchmark:compare against the branch report throws on the pass count;
v1.78.0 reads 2, the branch reads 1.
Issue: #743
User-Visible: no
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
The body of _process.yml is read from dev (@dev, #623), so the flags and
formats it passes to scripts are dev's. After "Опубликовать документ ревью"
the working copy of job integrate is the task branch, and a show/ship branch
with a clean merge is not rebased before review: its scripts/ may lag dev by
days. review-doc-guard.mjs silently ignores unknown flags (the anchor lost
#726 route and #737 usage), and a stale merge-candidate.mjs merges the old
way. Only two calls (#723 push refusal, #726 route) were taken from dev, each
with its own extraction, and on ship/reuse the remaining ones ran dev's
version anyway: the script version depended on the path.
Now one step right after setup-node extracts
`git archive origin/dev scripts .github/workflows/validate.yml` into
$RUNNER_TEMP/dev-tools and every repo script of the job runs from there via
TOOLS (review-result-gate, review-doc-guard, reviews-index, merge-candidate,
process-track route, status-label). validate.yml is part of the snapshot
because workflow-jobs.mjs reads it relative to itself; without it ci-proof
answers `failed (#622)` and every code merge would return to S6. The working
copy stays the material: git, the document and paths are judged there.
PROCESS.md §10.4 gets the paragraph "Скрипты конвейера — из dev": the
model_review exception, merges of pipeline changes judged by dev's version,
and compatible edits of the Validate proof contract.
Tests: test/process-integrate-tools.test.mjs is the job contract (no step
calls scripts/ from the working copy, every call goes through the snapshot,
one archive from origin/dev with validate.yml, and the step as is yields a
directory where ci-proof resolves the job contract); publish-push-refusal
runs the publish step and the #413 step on real bash with a task branch whose
review-doc-guard.mjs exits 7 (red with the old call). Existing harnesses take
the snapshot step before the publish and decide steps; the #706 mutant anchor
follows the status-label call.
Issue: #749
User-Visible: no
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
The weekly report disagreed with its own definitions (#728) and with the
pipeline texts that appeared after it (#705, #723, #729).
Volume. A task's volume counted documentation, so the archive move of #682
weighed 333 060 lines and #680/#681 thousands, all landing in the "> 1000"
bucket and shifting the cohort medians; and every commit with a Release:
trailer was dropped, including the task's own golden acceptance and test
re-pinning commits. Volume is now the +/- of class A and B files only.
Only beta commits are left out: the beta or release candidate (Release:
trailer plus the candidate subject or a changed bundle, bundle-policy.mjs,
drops every Release: commit, so it is not reused. A task whose commits
touch only docs/reviews/** has no volume and is no longer read as
infrastructure: the pipeline writes those documents, not the task.
Return reasons. Every non-merge outcome of merge-candidate.mjs fell to
"unknown". merge-candidate.mjs now exports a sign for the heading of each
outcome comment (OUTCOME_SIGNS; the step-failure text moved into
commentFor as 'error', byte for byte), and a test on the templates
themselves holds every case to its own sign. The report maps merge-stage
outcomes after a green verdict to "merge" and a push refused while
rebasing before review to the new "push-refused" reason.
Spec drafts. A new section after "По трекам" counts the S4-spec-review
epochs on the ask track for tasks whose first S5-ready falls in the
window, the epochs with a "Черновик:" comment (§7.2) and whether the draft
went to S5 or was thrown at S3, Spec-Draft: commits in dev for the window,
and S5 -> S7 per track for tasks with and without a draft, "мало данных"
under three. The snapshot also reads timelines of tasks in S5-S7.
Three #728 assertions pinned the old behaviour (a Release: fixture without
the candidate subject, and the rebase workflow refusal read as unknown);
they now expect the new definitions.
Issue: #752
User-Visible: no
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
`.room { transition: 0.12s }` interpolates a room's colour and its
fill-opacity / stroke-opacity independently, and the visible opacity is their
product. The states without a fill kept their alpha in the colour with the
default opacity 1 (.overlay transparent, .yard rgba 0.14, .outlined rgba
0.06 / 0.55, .picked rgba 0.25), while .styled writes an opaque colour plus
fill-opacity: var(--room-fill-op). On a change between the two on the same
node one half rose while the other fell, and mid-way the room was darker than
at either end. Opening the space settings on a floor with no fill (the dialog
shows "no fill" as its own colour at alpha 0) flashed every room grey for
~0.1 s, 0 -> 0.241 -> 0; cancelling the dialog after a preview, entering and
leaving the plan editor briefly darkened the fill (0.18 -> 0.317 -> 0.06).
Every .room state now writes an opaque colour plus *-opacity, and
transparent only together with a zero opacity. The transition itself,
.styled and the --room-* variables are unchanged; the space card takes the
same styles. The resting paint is the same: the witness records each state's
colour and visible opacity as dev drew them, and screenshots of seven resting
states (View without fill, with fill and borders, plan editor, room picked for
a merge, yard with and without borders, yard in the plan editor) are
pixel-identical to dev outside the plan editor's tool hint, whose text shifts
by a sub-pixel between runs on dev too.
Witness: new demo/smoke_room_fill_transitions.mjs, deterministic. A
MutationObserver pauses the room's transitions at their first frame right
after Lit commits, and the smoke seeks them through 0..120 ms in 15 ms steps.
Red on dev: fill overshoot 0.241 / 0.125 / 0.137 / 0.134 on the four paths,
stroke 0.241 / 0.242 on the first two. A real colour change of the same room
still runs a fill transition (catches `transition: none`).
A card-mod rule that sets only `fill` on an unfilled room now meets
fill-opacity 0; CHANGELOG and STYLING-HOOKS say to set the opacity with the
colour (the values are generated, STYLING-HOOKS §3.3).
Issue: #746
User-Visible: yes
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
tokenUsage summed the usage line of every review document in HEAD: the
report had no window, and every week repeated the whole history.
A document now enters a week's tokens when the commit that added it to
dev falls in [since, until]. fetchSnapshot reads the committer date from
git log -M --diff-filter=AR over docs/reviews and legacy/reviews:
an add sets the date, a rename (the #682 archive move) carries it to the
new path instead of adding the document again. The "missing" count of
#737 (hp:usage-none) follows the same window. ship findings keep reading
every SHIP-REVIEW document; only the token sum is windowed. Without the
date map (a unit over ready documents) there is no window, as before.
Proof is a temporary git repository with dated commits: a document
outside the window, one inside, an hp:usage-none pair on both sides and
an archive move inside the window; only the inside documents count.
Issue: #761
User-Visible: no
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
No workflow sets `shell:`, and GitHub runs such a step as `bash -e {0}`,
without pipefail: the exit code of `… | tee` is tee's, and a failing left
side passed silently. Three steps were unprotected:
- _process-resume.yml: an exception of process-resume.mjs (gh, API) left the
step green and the resume event was lost until process-reconcile;
- release-review.yml: a failed `prepare` went on with an incomplete
GITHUB_OUTPUT and proceed=true;
- validate.yml: a failed `classify-changes.mjs --heavy` left `heavy` empty,
heavy jobs were skipped and job `changes` stayed green.
Each gets `set -o pipefail` as the first line of `run` (validate.yml's step
becomes a block), following #727 and #472. test/workflow-pipefail.test.mjs
walks every .github/workflows/*.yml: a `| tee` line in `run` must follow
`set -[a-z]*o pipefail` or the step must have `shell: bash`; on the old tree
it names exactly the three places, and the _process-resume and validate
steps run on real bash under `bash -e` with a failing node.
ci-proof.mjs exports githubApiBase(env) (GITHUB_API_URL or
https://api.github.com, no trailing slash); githubCandidateTree,
loadGithubProofContext and release-gate's workflowRunsUrl take `apiBase`
with that default instead of the hardcoded host. night-red.mjs passes the
base directly and drops the fetch wrapper that rewrote the prefix. On
github.com the runner's GITHUB_API_URL is the same host, so behaviour there
does not change; archive_download_url stays as the API returned it.
The `mode` input for ship-review is out of scope (thin file in main, #716).
Thin files are not touched: _process-resume.yml is a body, validate.yml and
release-review.yml are not thin.
Issue: #751
User-Visible: no
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
The code-review prompt sat at exactly 1 400 of its 1 400 words (#634), so
any new line turned the budget test red, and #707/#726 already had to route
their notes through job outputs. Part of the text was dead or a retelling
of the reviewer digest, which #634 says the prompt must not repeat:
- the mutants fragment of the track line: since #709 `mutants` is always
false, so «прогнаны Validate» was unreachable;
- «Отсутствие мутантов по диффу — не находка» in the show line: a rule of
every track, already in REVIEWER.md «Трек show» and §10.4;
- three retellings — the repeated round, the gate scope and the severity
paragraph — now one-line references to the REVIEWER.md sections. The ban
on a separate issue for an in-scope Medium stays in the prompt: the model
files issues itself, and that mistake is expensive.
What only the prompt said moves into REVIEWER.md with links to the canon:
the spec delta is the diff of the issue body, a doubt about locality means
a full review with a stated reason, the three smoke-select answers
(docs/TESTING.md), and geometry without invariants in the report is an
unrun gate.
The prompt is now 1 130 words; the 1 400 threshold stays, the difference
is headroom. A new test ties every «docs/process/REVIEWER.md, «X»»
reference in the prompt to an existing `## X` section.
Issue: #750
User-Visible: no
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
Rule 8 skipped the status check for any range without class A files, so a
task in S3-spec or S4-spec-review could push a branch of tests, demo or
scripts with only a warning. §11.8 forbids exactly that: before S5 neither
class A commits nor the branch itself is pushed, because the spec-review
step takes the freshest origin/issue/<NN>-* as its material and lays the
SPEC-REVIEW document there, putting code in front of a spec reviewer who
must not read it (§2.4).
The #562 entry was written for a task before its first S status. The
decision is now made per issue in checkIssueStatuses: the status stays
optional only when the range is infrastructural and the issue carries
neither S3-spec nor S4-spec-review. Such an issue gets a rule 8 refusal
naming its status and §11.8; the range-wide #562 warning is still printed.
No status, S1-new/S2-analysis (reviewer-filed infra issues) and S5-S8 keep
their old outcome; closed, blocked and fail-closed checks are untouched;
rule 10 is still called only for ranges with class A.
PROCESS.md §10.2 gets the one-sentence exception, the mutation registry a
mutant that drops the S3/S4 condition.
Issue: #753
User-Visible: no
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
A same-route remount that cannot edit yet - hass arrives after the
element is inserted (the demo's own order), or a non-admin waits for the
server's can_write - keeps the editor in _pendingNavMode and enters it
later through _resumePendingNavMode -> _setMode. The draft revival was
wired only into the immediate warm adoption (_requestMode(..., adopt)):
on the pending path the draft was lost, _warmRevivePending stayed up for
the instance's life, _warmSnapshot stopped writing dlg, and the next
remount brought back the predecessor's stale draft.
The adoption tail (draft revival once under a held refit, then the
settled stage as the refit baseline) is shared by both paths: the
sequencing lives in src/warm-mode-adoption.ts, the refit bookkeeping in
the card's _holdWarmRefit/_releaseWarmRefit. The pending mode still enters through _setMode, the transition
authority smoke_nav_persist holds it to; resumeWarmMode then settles the
revival. _setMode ends the passive boot grace and refits the camera to
a header measured before the editor chrome rendered, so a camera the
pending window left untouched is put back and held exactly as an
immediate adoption holds it; a camera that has already moved on (View
refit, the user's pan, another space) is left to the ordinary refit. A
mode that did not commit, an explicit mode command in the pending window
and a route departure settle the revival too: no outcome leaves
_warmRevivePending up. The core file gives back 4 lines.
smoke_warm_dialogs gains section H through the UI: the three late-write
orders keep mode, draft, dirty baseline and a frame-by-frame identical
viewport; the chain carries this instance's draft, not the predecessor's;
a space switch in the pending window eats the draft. 14 checks are red
on dev. The mutant warm-pending-mode-leaves-revive-waiting is guarded by
the smoke; docs/WARM-REMOUNT.md §2 describes the pending editor.
Issue: #756
User-Visible: yes
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
The pipeline dispatches a full Validate for a `ci:golden` task, and
`screenshotsGateMode` read `full=true` as strict on any ref. Between betas
the source fingerprint on dev is legitimately stale (#479: re-captured for
the beta candidate), so every visual task failed preflight on the
conveyor's material until its author re-ran `docs:accept --identical` on
the current dev - #718 and #740 both did, and two visual tasks in a row
could not merge without it. On a task branch the mode is now `warn` even
with `full=true`; dev, main, PRs, the schedule and Release: candidates stay
strict.
Issue: #760
User-Visible: no
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
Since #735 switchCycleMs times the warmed twelve-switch cycle, and the
absolute ceilings of 7000 ms (flat) and 8000 ms (2.5D) sat 7.6-10.2
times above the level. performance_smoke judges only these ceilings, so
between full runs the warm floor switch that #694/#725 just sped up was
guarded only against a several-fold collapse.
Series: every Full Performance run after #735, both sides (the base is
measured by the candidate runner, so it is warm too), 7 samples each -
36821241343 (#735, base 76558bf2), 36838891001 (#740), 36838952536
(#742) and 36839009721 (#739), the last three against dev 7ff2b5ae.
flat large-house-v1 / plan-snap / interaction 666.9-812.7 ms
2.5D large-house-isometric / stage3-dense 766.5-1333.9 ms
The 2.5D maximum is the dense pair of 36838952536, whose base on the
same runner read 1249.7 ms against 849.9-982.4 ms elsewhere: runner
noise the series is meant to contain. No 3-sample performance_smoke
median is in the series yet; those profiles join Validate only on a
src/** diff.
Rule (as #692, #675 falls in the same band): one number per family, the
first multiple of 50 ms at or above 1.15 x M and no higher than 1.2 x M,
M being the family's maximum median: flat 1.15 x 812.7 = 934.6 -> 950
(+16.9 %), 2.5D 1.15 x 1333.9 = 1534.0 -> 1550 (+16.2 %). One number
per family keeps the smoke = full (#473 AC4), plan-snap/interaction
"every original ceiling" and dense = historical (#160) contracts; the
price is wider headroom for the faster profiles. The base-relative
comparison of the full workflow (0.35 / 0.2, 250 ms) is unchanged and
stays the detector for smaller growth.
The new test pins both families: one ceiling in every file of a family,
every point of the series passes the smoke budget with --absolute-only,
the ceiling follows the rule and stays inside [1.15, 1.2] x M, doubling
the level fails, and the full profiles' ratio and noise allowance are
unchanged. 7000 left in any flat file or a ceiling under 1.15 x M reds
it. The README records the series and the reasoning.
Issue: #747
User-Visible: no
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
The space card drew its rooms with a bare map(), so Lit reused room nodes by
position, and `.room { transition: 0.12s }` (planStyles is part of this card's
styles too) drew a node's fill and stroke in from whichever room held it
before. Two paths change the room set in the same DOM: a new `space` in
setConfig of the same element (the card editor's preview), and a config event
from any device that inserts, removes, reorders or re-zones a room of the
shown space. Filled rooms faded out and back in for ~0.12 s, unfilled ones
briefly darkened in a filled room's place.
The list is now keyed(space.id, repeat(rooms, (r, i) => r.id || i, ...)),
the same shape as the full card after #742: the outer key handles the space
change, the inner one keeps a node bound to its room inside a space. An
id-less room keys by its numeric index, which never equals a string id. The
transition itself stays: it smooths a real fill change on the same room. The
#742 note in plan.styles.ts now names the space card as well.
Witness: a new section of smoke_space_card. The config is delivered by a
server push (__hpTest.setServerConfig), the event the card subscribes to.
Red on dev: node r1 reused for g1 with fill/fill-opacity transitions and a
first-frame fill of rgba(0, 0, 0, 0) / 0; a room inserted first shifts all
four nodes and replays fill transitions. A real custom_fill change still runs
a fill transition on the same node (catches `transition: none`). One mutant:
inner key replaced by map(), guarded by AC2 (checked by hand: red).
Issue: #745
User-Visible: yes
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
Three independent blind spots in the test harness.
1. smoke-select read symbols only from changed lines of a --unified=0
diff. An edit to the arguments of a multi-line call names nothing:
#741 (d5bdfde9) changed only the arguments of
runtime.resolveIsoOverlayFitEnvelope({ on the line above, and the
selection answered "unproven" plus the visual minimum, although the
callee is registered in smoke-links for smoke_iso_flat_parity and
smoke_isometric_contract - the two smokes the #741 author ran by hand.
The selection diff now carries CALL_CONTEXT_LINES = 3 lines of
context; for each changed line parseDiff looks for the nearest
unclosed "(" above it within the hunk, walking through a literal
argument ({ or [ after "(", "," or "["), stopping at ";" on depth zero
or any other unclosed brace. A callee from the symbol table joins
symbols and the new callees field and is marked "(вызов)" in the
report. Context lines never give direct symbols. task-packet takes a
separate context diff for selectSmokes; change-risk keeps --unified=0.
Over the last 80 src commits of dev: 16 commits gain a callee, 2 move
from unproven to a proven link (#741, #7245f8e8ca7), +15 smokes in
total, at most 4 per commit, none lost.
2. The #732 dead-field check judged only scene-builder calls. The four
resolveIsoOverlayFitEnvelope({...}) literals in iso-scene-render tests
went straight into the test-build function, so stageSize: null (the
field #741 removed) stayed green. They now go through overlayFit typed
with OverlayFitFixture (keys of IsoOverlayFitEnvelopeInput); the check
judges overlayFit/resolveIsoOverlayFitEnvelope calls like the scene
builders, and its probe asserts that OverlayFitFixture rejects
stageSize, so the type resolved to the real input and not to any.
3. smoke_backdrop's mode() called the private _setMode and slept 220 ms.
It now enters a mode through __hpTest.setMode and waits for the end of
the transition by the same markers as section 6b (#715): one page
helper used by both. Oracles and the 59 check names are unchanged.
Witnesses: d5bdfde9 selects both iso smokes with no "unproven"; the same
fixture without context lines is unproven again; attribution disabled
reds both AC1 units. stageSize: null in an overlayFit call reds the first
#732 test; a direct resolveIsoOverlayFitEnvelope({...}) reds the third.
smoke_backdrop is green normally and with animation frames slowed to 60
and 150 ms; a stage animation that never ends fails with a named error.
Issue: #754
User-Visible: no
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
The risk table of #707 judged every non-comment line of a class A file as
code. On the history since 15.09 that raised #741 from ship to show for a
removed interface member that never reaches JS, and put false classes on
#624 (removed imports), #693/#694 (stairs-view is rendering, not geometry)
and #725 (the config fingerprint memo is not the config schema).
- Lines of module syntax and TypeScript types give no risk, like comments:
`import …`, `export … from …`, `export type …`, the head of `interface X`
or `type X =`, and the lines inside such a block (indented, plus the
closing line). The block state per side of a change block starts from the
hunk context git writes after `@@ … @@` and follows every unindented line
of the block, so a member under `@@ … @@ export interface X {` and a whole
interface added in one hunk are judged alike. Only `.ts`, and not in the
`migration` area: there the types are the config contract (#588, #649).
- `stairs*` is narrowed to the stairs model (`stairs`, `stairs-box`,
`stairs-editor-model`); `config-*` to writing and adopting the config
(`config-adoption`, `config-store`, `config-reload-authority`,
`config-write-conflict`).
- A replaced line is one piece of evidence: a removed line whose counterpart
in the same change block hits the same class is folded into it instead of
printing `path:N (удалена)` next to `path:N`.
classifyRisk stays a pure function over the diff text. On the history the
raising classes change for #741 (none), #693 (visual only), #694 (no
geometry), #725 (perf only) and #624 (no devices/perf); migration on #588,
#612, #649 and #661 stays. PROCESS.md §5 names the new exemption.
Issue: #755
User-Visible: no
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018qZfe7YS4rqEMKoVeS3GKd
