Share-link integers and Cesium radians were converted before normalizeBox
checked them, so a blank edge still became 0°. Edges are now accepted only as
numbers or non-blank numeric strings, which also refuses [] and {}. Tests cover
all four edges through validateBox, dequantizeBox, boxFromRectangle and
searchHls.
Adds a case that fails before de-duplication (a 24-slot overview held repeats
and reported itself truncated) and says why the date-line halves overlap.
The existing "route that fits" case used 199 steps, which is how the
off-by-one went unnoticed. Folds the docs note into the existing cap sentence
and makes the changelog entry specific.
Preserve upstream changes, restore the reviewed Radio ownership regression,
and retain reporter raysung-2026 credit and #804 provenance.
Reconciles upstream main 81cf437 and PR #816 head 702ecc3; no force push.
Keep this branch's CHANGELOG entry above upstream's; upstream main no longer lists the GBFS station-name entry an earlier merge carried here.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Seven QA scripts hard-coded macOS Google Chrome. They now share chromeExecutablePath(): PUPPETEER_EXECUTABLE_PATH when set, then macOS Chrome when installed, otherwise Puppeteer's managed browser. The cables screenshot path is built with fileURLToPath so it works on Windows. qa-overlay-baseline keeps its fixed Chrome because its baselines record the executable.
Co-authored-by: Sameh Khamis <[email protected]>
Number(null) is 0, so the cockpit cloud profile treated a missing weather code
as clear but available. The briefing's three checks now share one helper, and
the current-state note is its own bullet instead of folding into the route one.
A single % made the seek delta negative when the target angle was more than
540° below the start, so 350° to -350° turned 340° through 180°. Uses the
same double modulo as playback, and adds tests for exact ±180° endpoints and
a near-antipodal seek.
Clears the timeout guards once each race settles, measures the 15,000-vehicle transit ingestion with main-thread CPU time instead of wall time, and times the region-timeout case from the boundary lookup with a 2 s ceiling, below the 3 s default budget, so an ignored budget still fails.
Co-authored-by: Sameh Khamis <[email protected]>
The test's access stub matched a forward-slash suffix, so on Windows the resolver's backslash path never matched and it fell back to PATH. The stub and the expected value now use the same path.join result. The Windows CI job also runs this test file now.
Co-authored-by: Sameh Khamis <[email protected]>
The horizontal offset multiplied elapsed time by a sine, so flakes swung faster the longer the app ran. It is now the integral of the intended sway velocity, which stays bounded at any session age; Wind still scales the amplitude.
Co-authored-by: Sameh Khamis <[email protected]>
Exclusive presentation hides collapsed right-rail panels with display: none.
Skipping every display: none panel also skipped those, so panels the rail had
auto-collapsed stayed collapsed after the HUD was hidden or left Tactical.
Only panels hidden by other CSS are skipped now. Adds the changelog and
current-state notes.
- Pin finiteCoordinate helper behavior across numeric, non-finite, and string inputs
- Verify recovery stage downrange distance, flight numbers, and coordinates fall back safely to null on non-finite data
- Pin recovery status transitions for LOST and EXPENDABLE stage categories
- Verify payload massKg and amount normalization edge cases for blank, negative, and non-numeric values
- Guard against legacy pack bootstrap during unreadable storage recovery
- Report failed saves honestly across captures, updates, and imports
- Resolve default scene id correctly when advancing past a specific shot
- Count unique scenes executed rather than queued shots in run telemetry
- Return run failure status when playback throws instead of false success
- Pin five regression scenarios in src/scenes/director971.test.mjs
Rudy Celekli