The cursor only changes focus now, and the rows no longer measure themselves
(no BoxWithConstraints per row). Measured upstairs before: 75 ms a press up/down.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It used the machine's own sans-serif: Noto Sans here, the wider DejaVu in the CI
container, so 8 checks failed only in CI. Noto Sans (OFL) is now a test resource.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Android's dynamics processor on the player's own sound session: quiet parts lifted
a little, loud parts held back, a limiter on top. Overnight is 23:00-07:00 and
starts and stops by itself. Off until Paul has listened to it; set from adb
(orbit_even_volume off|overnight|always).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PlaybackInfo's RealFrameRate (or AverageFrameRate) is carried in the plan, and the
provider contract gains an optional frameRate that every provider can fill (Plex
and local files later); without it the first frames are measured as before. Full
screen is a dialog, so the preference is set on the dialog's window as well as
the activity's. Still off unless the frame-rate switch is on.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
What's taken off Continue watching (by hand, or watched to the end) is recorded
in its own gone-continue file for three days and passed on by Rooms sync, so the
other TV's copy can't bring it back; watching it again later does. A Taste reset
in either room wins over older events. Older Orbits ignore the new file.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Unknown fields on a known signal are carried through; a line with an unknown kind
or schema is kept exactly as written (if it has an id and a time). Sending, merging
and the start-up tidy all keep them. Once both TVs have this, a new kind of
signal (Taste Stage 2's caught-end) can be installed one TV at a time safely.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Whoever was watching is highlighted; OK, Back or 5 seconds carries on as them,
so it costs nothing when it's the same person. Choosing someone else follows the
usual rules (PIN for Adult, Kids locks). Only when Orbit asks at all. The press
that brings it up does nothing else. A full-screen programme isn't interrupted.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Paul's Emby provider allows 3 streams across the house. Ahead of the ident only
the next show's details are asked for (IsPlayback=false, no stream); its player
is created once the ident covers the screen and the show under it has faded and
stopped (0.8 s). Neighbour channel look-ups are information-only too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Found by the on-TV playtest matrix: no test clip states its frame rate, refilling
after a seek counted as a stall (would stop the overnight comparison after three
tunes), and DivX 3 AVIs played as sound alone. Third matrix run: every case as
intended, including 23.976/24/50 Hz switching and the TV returning to 59.94.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The report (positionReached, every mismatch position, the runs so far) is saved
safely after each compared page. orbit_shadow_resume picks up from the saved
position; a start anywhere else, a resume without a report, or one after the end
is refused and noted in resume-refused.txt. verify.sh now builds clean: incremental
builds had left stale Compose stability checks in two earlier APKs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tv.orbit.classic.playtest has its own data and Orbit's own screens switched off,
so it can't reach the household's schedules, Taste or Continue watching. It plays
only the ten approved clips (checked by SHA-256) through Orbit's own player set-up,
container rule and frame-rate matcher, with each switch set per run. Each run
covers seeking, reopening (recovery), one retry on error, and the no-decoder case.
verify.sh now fails if the hook ever reaches the release APK.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
At least 1.5 s between pages (5 s while a programme plays), never faster than
the server answers. A failed page is retried once after 30 s. It stops by itself
if Orbit closes, playback stalls 3 times, memory passes 80%, or the server takes
over 20 s a page 3 times running. Stop and resume from adb. The report is written
every 25 pages so a stop keeps the results. Rebuilt into 0.123.0 (never installed).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Frame-rate matching sets the TV's display mode to a whole multiple of the
programme's frame rate (23.976 film at 23.976 Hz, 25 fps at 50 Hz), keeping the
resolution, and hands back to the TV's own choice when the programme closes.
AVI joins the direct-play containers only when its switch is on. Both are set
from adb (orbit_frame_rate, orbit_avi) and default off, so 0.123.0 behaves as
0.122.1 until then. Tests use the downstairs TV's real mode list.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
All 12 contract checks run against a real Silo (build-2072). 11 pass; pictures fail
because Silo redirects image requests, which Orbit never follows. Recorded as the
first need of a Silo adapter, with the identity question. The live test is skipped
unless a Silo server is set, so the release gate is unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The snapshot comparison moves into compare_snapshots.py with offline tests,
run by verify.sh. It agrees with today's two hand-checked installs.
It refuses to install while Orbit is playing, and keeps the last three backups per room.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Captured by the read-only shadow comparison on downstairs (9 November 2026, 5 pages, identical, catalogue untouched),
sanitised on the TV (titles, original titles and tags replaced, external id digits swapped, structure kept). Today's
reader and the provider path read them identically, page by page and as a whole import. Branch from the installed 314
build; no version change, nothing installed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- The Originals row scrolls itself: the focused card whole on screen at once (any running scroll
dropped), the peek easing in over 140 ms; rapid or held presses and Reduce motion scroll at once.
RowScrollTest checks the card is whole after every rapid press both ways.
- Asteroid Alley's rock: lumpy angular asteroids in three shapes and rotations, off-centre
craters of different sizes, a darker rim (Paul: the round one with a dot looked wrong). The
shield pickup is a shield shape, Double Stars two stars, the Forbidden gate three marks.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Five pilots from the cast library, each with an ability charged by stars (Ollie Patch Up, Blaze
Speed Burst, Zip Warp Ahead, Drift Slow-mo, Luna Moon Shield), chosen on the title
- Four regions in order with their own art: Asteroid Alley, Broadcast Boulevard, The Static Zone
(interference bands), The Forbidden Frequency with the 999 portal to a secret route and the wizard
- Star Magnet and Double Stars beside the Orbit Shield; region banners; furthest distance kept
- Gentle: one hazard a row everywhere, a fourth bar of hull, earlier warnings, calmer effects
- Four SoundCraft sounds of its own: boost, slow-mo, portal, region
- Bot dodge checks for every region x difficulty x pilot; 0 bytes allocated; prototype label gone
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Scores ride the Rooms sync beside Continue watching: each room sends its people's own runs and
counts; what arrives is kept as a read-only copy per room (arcade/rooms/<link>.json), merged only
by adding (same run once, best 10, counts only up). No save file is ever written by a sync.
- A person's scores from every room show as theirs: cards, the best to beat, and the board, now
"the whole house": Your best and House best with a name, and a house Top 10 with names
- Orbit Rush: shadow, size, an eased height and a coloured up/down marker with a time bar
- Originals row: room at both ends and above and below, a peek of the next card, eased
scroll that keeps the focused card whole (a jump with Reduce motion)
- Tests: RoomScoresTest, Rush lift, the row's fit; 1,095 pass
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- design/arcade/sounds: each sound a SoundCraft session (editable .scraft kept), soft sine and
triangle patches, filtered and levelled, delivered as small mono Ogg (133 KB for the kit)
- ArcadeSfx: the kit, each sound's level, most-at-once, gap and priority; SfxLimiter (the oldest
gives way); ArcadeAudio.sfx() reads the settings at most every 250 ms, so playing allocates nothing
- Originals: Orbi-Raiders (new boss-hit and damaged events), Orbiteroids, Channel Chaos, Orbit Rush,
Night Shift, Test Card Tangle, Codebreaker, Quiz Night
- Classics: Block Drop, Blast Off, Skydoku, Comet Trail, Brick Break, Pairs, Patience (GameSounds
hears each change of the game's value)
- A Sound effects setting for every game; Effects, Mute, Orbit Quiet and Wind Down still apply
- SoundKitTest: files present and small, lists cover the kit, caps and gaps, 0 bytes allocated
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Six more backdrops: Seaside Pier, Drive-In, Snowfall, Sakura Night, Retro Living Room, Moon Base
- Fit checks for the new games' buttons and every backdrop name
- docs/new-originals.md
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- General knowledge, movies and TV; Mixed or one category from the Mode button; five rounds, turns, no timer
- Answers 2x2 with the D-pad and OK; results wait for OK; Back pauses; solo scores in High Scores, group games keep none
- Originals row on the Arcade home scrolls when there are more than five
- QuizNightTest and fit checks for every question and answer
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Arcade buttons with a value on the right: the label takes all the room left (a spacer shared it, so "Sound" read "S…")
- Pause menu 250 dp, game settings sheet 380 dp, so every line and value fits at the largest font scale
- BumbleHive: Mode moved into the stack under Play, same size, top-to-bottom focus
- Fit tests: every game's settings, How to Play heading and the pause menu's sound line
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Sectors of rocks, tellies, satellites, interference clouds, a rogue beacon and the cosmic anomaly; Drift to find
- Hull and shield bars; title, play, sector card (waits for OK) and shared result from Stage B frames 50-53
- OrbiteroidsSimTest: rules, a three-player balance bot on each difficulty, 0 bytes over 10,000 steps
- All five Originals now play
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- 7x5 night sky: Owl hops, OK turns the beam piece; light the mast; clouds drift a tile a hop; moves against Owl's par, no timer, no fail
- Title, chapter select, play and level complete from Stage B frames 40-43; shared pause; Back by context
- Per-profile progress (best stars and moves per level); High Scores keep total stars; Kids-suitable
- Levels from design/night-shift/make_levels.py, checked again by a bot in NightShiftLevelsTest
- Power-ups left out for Paul's call (they need the Menu key the remote lacks)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Blast Off: twenty numbered boards with line targets and shape limits, same shapes each try, all beaten by a test bot
- Mode button shows the whole mode name; title buttons 230 dp, Play Again 180 dp
- Every mode its own High Scores table; Classic tables, kinds and Household keys unchanged
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>