Network failure of the editor runtime is no longer terminal: the loader
re-arms to idle and the next explicit press starts a fresh cycle, while a
fingerprint mismatch on either attempt stays terminal. The toast now says
what actually helps — retry advice for the network, refresh advice for a
foreign build — via one shared lazyLoadFailureMessage helper (new i18n key
editor.retry_advice in en/ru/de).
The field smoke caught a second, deeper bug on the way: Chromium records a
FAILED module in the page module map permanently, so retrying the same URL
(even the cache-busted one) never touched the network again. Every retry
now carries a per-cycle nonce and becomes a genuinely new module request.
A proxy-cached stale entry no longer kills the card silently: the entry
facade is rewritten at build time from a static re-export into a top-level
`try{await import(...)}catch{...}` — importers keep the happy-path
guarantee (await import(entry) still resolves only after
customElements.define), and the catch defines a fallback element with a
localized "reload the page" panel. Content-hashed chunks are served with
`public, max-age=31536000, immutable`, and verifyBundleTree now fails on
orphan chunks that the manifest does not name.
Proofs: loader units for re-arm/terminality/toast wording + an AST check
that both loaders forward the terminality flag; smoke_entry_stale (en/ru)
against a tree without the main chunk; smoke_lazy_editor_chunk extended —
second press after network failure now really opens the editor; pytest for
the immutable header; orphan-tree unit; five new registry mutants.
TESTING.md budget line updated to the #352 ceiling alongside.
Issue: #353
User-Visible: yes
Add Deutsch across all card surfaces and backend flows, backed by the language registry introduced in #62. German loads as a fingerprint-checked page-shared locale chunk so EN/RU remain synchronous and the initial View budget stays intact. Root render gates prevent mixed-language flashes, retry once, and fail open to English. Extend parity, runtime, bundle, browser and visual coverage, plus contributor and user documentation.
Issue: #348
User-Visible: yes
smoke-select flagged quantizeKeyCoord/INCIDENT_EPS/KEY_FACTOR as symbols no
smoke names — true by design: the smoke proves the verdicts through the
rendered card, never touching the internals that decide which nodes are one
node. The registry entry records that non-textual link (#241 rule).
Issue: #331
User-Visible: no
Все прежние гейты проверяют снимок модели. Дефекты геометрии рождаются в
редактировании: #289, #296 и #298 прошли решётку, кладку, роли, ключи и аудит
ручек, потому что такая геометрия снимок не портит — она портит следующий жест.
demo/smoke_edit_walk.mjs расшатывает реальный план продуктовыми жестами
(_rszEdgeDown/_rszMove/_rszUp, _confirmRoomDelete, optimizePlans) по фиксированному
семени и после каждого шага судит конфиг в node. Второго представления редактора
не появляется — принцип #292.
Подшаговый шум остаётся наблюдением, а не нарушением: координата пишется девятью
знаками, 304/240 = 1.266666667, отклонение 8e-8 шага неустранимо форматом
хранения и уйдёт на этапе 1 ADR #282. Судится только «вне сетки».
Таблица KNOWN работает в обе стороны: обход падает и когда находок больше, и
когда меньше. Молча позеленевший гейт не сообщает о починке — так
partition-mt2on9ou-0 прожил в плане владельца от беты 9 до rc.1.
Новый инвариант checkHiddenObstacles: перегородка на стене комнаты, незакрытый
контур на стене комнаты, черновик, который не может стать комнатой (#296).
Найдено сразу, в пяти прогонах из шести — на первом жесте: #298 (ресайз уводит
конец записи толщины мимо решётки и мимо ребра), #299 («Оптимизировать» и
удаление комнаты сливают записи через границу роли).
Issue: #297
User-Visible: no
Eight closed issues on wall junctions — #271, #272, #275, #276, #277, #278,
\#279, #280 — shipped in beta.9, and the break in the owner's real plan
survived. Every one of them was accepted on synthetic fixtures: a cell-5
mixed-depth T, a rectilinear T with three equal half-depths. On those the fixes
work.
The smoke asks the product itself, through isPointInFill on the wall path,
whether masonry exists where the model promises it. That is independent of both
resolution and the pixel-diff thresholds which miss this class: a 45-step break
on a large plan is a fraction of a per cent of the frame, well under the 0.05
per cent scene tolerance.
Measured on the shipped code: four breaks, 181 grid steps in total, 45.25 each.
That number is derivable rather than incidental — the node joins a 30 cm
exterior wall, a 30 cm spur and an arm five steps long; half-depth 15,
MITRE_LIMIT 4, corridor radius 60, and 60 − 15 = 45 steps are cut out of the
neighbouring 20 cm wall. The corridor eats the masonry, which is what #271 and
\#275 describe.
The fixture is privacy-minimised — neutral room names and ids, no device
bindings — and keeps its coordinates, because they are the point. It lives in
test/fixtures rather than demo/fixtures on purpose: sourceFingerprint hashes the
.mjs of demo/fixtures and demo/golden, so anything added there staleizes the
committed bundle and the screenshot manifest at once. Verified after this
commit: bundle still fresh, screenshots still current.
The debt is recorded as numbers and compared exactly. Better and the test asks
for the numbers to be updated, which proves the improvement; worse and it
catches the regression. Verified by execution in both directions.
Issue: #285
User-Visible: no
git grep _bindingCandidates -- test/ demo/ was empty: the function that decides
what the user is offered under Add had no test and no smoke. Tombstones were
covered from every side, the list they filter was never asked. That is how #262
reached us through a user report instead of a gate — smoke_hidden_flag assigns
binding straight into _markerDialog and bypasses the picker entirely.
Twelve checks against the real bundle: a deleted device is offered again, a
deleted plain entity is offered again behind the checkbox, the checkbox itself
is the trap (a device entity is absent with it off, present with it on, and it
starts off for a new marker), a placed binding is not duplicated, and re-adding
replaces the tombstone and leaves the picker.
The twelfth pins the known defect #262 as current behaviour: a device tombstone
still hides its child entities. Fixing it turns the check red and forces it to
be flipped, so the fix cannot pass the coverage by.
smoke-links registers only the pure tombstone helpers. The picker names itself
and is found by direct match; the helpers are not named anywhere in the
scenario. Verified by probe: touching src/devices.ts alone selects this smoke
as a registered link, and without the entry nothing would select it.
Issue: #263
User-Visible: no
