# BumbleHive native: verification (8 November 2026) Branch `arcade-bumblehive-native` from `arcade-orbi-raiders` (`760c957`, which also carries the power-up sheet change), worktree `/home/pauljd/Documents/OrbitTV/orbit-arcade`. Version unchanged (0.110.5-arcade-stage-c, 281): for review, not a release. Nothing pushed, merged or installed; no TV or adb used; `orbit-tv` and `orbit-tv-release` left alone. What it is: `docs/bumblehive-native.md`. ## Update: Hard's own High Scores (8 November 2026, second commit) Paul chose a separate High Scores table for Hard. Same Gradle run afterwards: **853 run, 24 skipped, 0 failed** (5 new in `HiveScoresTest`), Node 28 passed, lint 0 errors and 85 warnings (unchanged, no new kinds), release APK 47,793,763 bytes (unchanged to the byte), debug +82 bytes. `HiveScoresTest`: each mode reports only to its own table; a dozen Hard runs leave Normal's record and its JSON in the file identical, and Normal runs leave Hard's alone; a file from before reads with an empty Hard table; the row's switch, label and Top 10 title (every other game has one table and no switch, and `HiveHard` is not a game); Clear high scores clears both BumbleHive tables and nothing else. ## Checks (first commit) One Gradle run (`assembleDebug assembleRelease testDebugUnitTest lintDebug`, orbit-tv's `.tooling`), against `760c957` built the same way in a scratch worktree. | | arcade-orbi-raiders (760c957) | arcade-bumblehive-native | |---|---|---| | Build (debug and release) | passed | passed | | Kotlin unit tests | 815 run, 24 skipped, 0 failed | **848 run, 24 skipped, 0 failed** (33 new) | | Node tests (`tools/test-bumblehive.sh`) | 23 passed | **28 passed** (5 new) | | Lint | 0 errors, 67 warnings | **0 errors, 85 warnings** | | Release APK | 47,728,227 bytes | 47,793,763 bytes (**+65,536**, +0.14%) | | Debug APK | 52,019,816 bytes | 54,403,753 bytes (+2,383,937; the files inside grow 215 KB, the rest is how the incremental debug build packs its dex files) | Lint: no new warning kinds from the code. UseKtx 57 → 66 (nine `Canvas.withTranslation`-style suggestions in `HiveArt.kt`, kept as plain save/restore so the drawing makes no lambdas per frame). The other 9 new lines are lint's "newer version available" checks on `build.gradle.kts` and the Gradle wrapper (GradleDependency 6, NewerVersionAvailable 2, AndroidGradlePluginVersion 1): about the build files, not this change, and absent from the baseline run only because of lint's version lookup. ## Parity with game.js `BumbleHiveParityTest` against `src/test/resources/bumblehive/parity.json`, recorded from the real `game.js` under Node: - **200/200 levels identical**: every cell, colour count, rows, shape and panic time. - **469/469 scripted landings identical** in 15 levels (Honeybombs included): grid, score, ceiling, drops, panic, screen after each; one win and five swarms among them. - **26/26 shot paths identical frame by frame**, bounces included (largest difference 0.0). - `hive-parity.test.js` checks the record still matches `game.js`. ## New tests - `HiveSimTest` (20): Hard numbers (panic ×0.6 min 5 s, colours every 10, 3 swaps), Hard's layouts the same as Normal's, **Hard's save separate from Normal's** both ways, the save round trip and a level reset keeping Honeybombs; weighted aim, one shot per OK press, swaps and Honeybombs, the ceiling drop and swarm, win score, stars and unlock, burst and fall bonus, runs reported once however they end, Back, Level Select; Daily Bonus (free, Not today, once a day, no streaks), Reduce motion's wheel; help toggles; the Household question (copied not linked, asked once, waits until answered and merges); **allocation: 10,000 ticks of bot play, 438 shots, 0 bytes allocated in the frames**. - `HiveLegacyTest` (10): **the import equals what was stored**; runs once and never rewrites; no old save; empty and corrupt old saves kept as they were but not offered; **a failed read writes nothing and is retried**; an unreadable snapshot isn't replaced; old per-profile keys recorded; the read script only reads; continuing copies and leaves the snapshot alone (a second profile can too); the new `kept` field survives the file and a reset, old files still read. - `BumbleHiveParityTest` (3), above. - Node (5): the record matches game.js; **the real READ_SCRIPT against a storage that refuses writes: the old save returned exactly, nothing written**; missing, empty and unreadable storage; the import uses the old game's URL; game.js loaded with no bridge (if the WebView ever runs it) writes nothing. - Changed: `ArcadeSessionTest` and `ArcadeWordsTest` (BumbleHive's level progress can now be reset per profile). ## Not checked - **Nothing was seen running**: no emulator, no TV. Drawing, text and borrowed glyphs, focus, the result card, the tones, and frame times and memory on the 2 GB box are checked only by the compiler, the tests and reading. - **The import has not run on a real WebView.** Origin sameness is reasoned from the old code (same URL, settings, default data folder), not observed. - Drawing allocation couldn't be measured off the box (only the rules were). ## Proposed check on the box (for Paul to approve; not run) Before install, read-only: snapshot the old save so it can be compared after. ```bash ADB=/opt/android-sdk/platform-tools/adb; D=192.168.0.252:5555; P=tv.orbit.classic; OUT=~/Documents/OrbitTV/audits/bumblehive-native-$(date +%Y%m%d-%H%M); mkdir -p "$OUT" $ADB connect $D $ADB -s $D shell run-as $P ls -la app_webview/Default/Local\ Storage/leveldb > "$OUT/leveldb-before.txt" ``` After install, once BumbleHive has opened on the title: check `files/arcade/household-bumblehive-legacy.json` shows 48 unlocked (the WebView itself may touch its LevelDB log files when it opens storage, as it did every time the old game opened; the save's key is only read), then `dumpsys gfxinfo` over two minutes of play as for Orbi-Raiders.