BackendConfigScreen missing LOCAL_WHISPER card — only 3 of 4 backends render despite source defining 4 in priorityOrder default #144

New issue

Closed

opened 2026-09-11 04:08:16 +00:00 by Mark · 1 comment

Mark commented

2026-09-11 04:08:16 +00:00

Owner

Bug

The Backend Configuration screen (Manage → Backend Configuration) is missing the LOCAL_WHISPER card from the backend priority list. Only 3 backends render: Android SpeechRecognizer, FireOS Voice Service, Cloud API. LOCAL_WHISPER does not appear.

Reproduced

Device: Fire HD KFTRWI (Amazon FireOS)
App: com.voicetranscriber v1.0 from 93b7fd1
After pm clear com.voicetranscriber (verified default-config state): still 3 cards, no LOCAL_WHISPER
AGENTS.md claims "multi-backend (ANDROID/FIREOS/LOCAL_WHISPER) shipped 2026-09-08 with LOCAL_WHISPER as primary"

Source code expectation

app/src/main/java/com/voicetranscriber/service/SpeechBackend.kt:62-67 defines the default priorityOrder as 4 entries:

data class BackendConfig(
    val priorityOrder: List<SpeechBackendType> = listOf(
        SpeechBackendType.ANDROID,
        SpeechBackendType.FIREOS,
        SpeechBackendType.CLOUD,
        SpeechBackendType.LOCAL_WHISPER
    ),

BackendConfig.load() at line 138-162 returns either the saved order or the default. After pm clear, the prefs are empty so order is empty and load() returns BackendConfig(settings = settings) — which uses the data class default of all 4. The screen should render 4 cards.

Models screen proves the type exists

The Models screen (Manage → Models) does correctly list Whisper models for download (Tiny, Tiny (English), Base) and renders an "Active model: No model installed" empty state. The LOCAL_WHISPER backend type is fully wired in code:

SpeechBackend.kt:20 — LOCAL_WHISPER("local_whisper") enum entry
TranscriptionService.kt:343 — handler
BackendConfigScreen.kt:546 — display name "Local Whisper"
BackendConfigScreen.kt:553 — description "On-device whisper.cpp model"

But none of these reach the priority list UI.

Impact

Users on FireOS cannot promote LOCAL_WHISPER to the top of the priority list, so the SpeechRecognizer failure on FireOS (see #143) cannot be worked around
The BackendConfigScreen shows only 3 backends even though the source says 4 — code/build drift or runtime filtering bug
The "primary" backend per AGENTS.md is unreachable through the UI

Suspected cause

Either (a) bytecode in the installed APK differs from the source tree (the source HAS 4 backends but the compiled APK has 3), or (b) the itemsIndexed(priorityOrder, key = { _, backend -> backend }) in BackendConfigScreen.kt:194 is dropping LOCAL_WHISPER at runtime, or (c) the encrypted prefs on this device had a saved order from a prior install that excluded LOCAL_WHISPER — but this was not reproducible after pm clear, so (c) is ruled out.

Suggested fix

Compare the installed APK bytecode against the source tree to determine which of the 3 backends is missing and why.
Add a regression test that asserts BackendConfig.default.priorityOrder.size == 4 AND that the screen renders 4 cards.
After the fix, ensure that LOCAL_WHISPER is the default FIRST backend in the priority order on FireOS (since the Alexa voice service binding fails on Fire devices — see #143).

Reported by Hermes on 2026-09-10 during smoke testing of v1.0 (build from 93b7fd1) on Fire HD KFTRWI.

## Bug The Backend Configuration screen (`Manage → Backend Configuration`) is missing the LOCAL_WHISPER card from the backend priority list. Only 3 backends render: Android SpeechRecognizer, FireOS Voice Service, Cloud API. LOCAL_WHISPER does not appear. ## Reproduced - Device: Fire HD KFTRWI (Amazon FireOS) - App: `com.voicetranscriber` v1.0 from `93b7fd1` - After `pm clear com.voicetranscriber` (verified default-config state): still 3 cards, no LOCAL_WHISPER - AGENTS.md claims "multi-backend (ANDROID/FIREOS/LOCAL_WHISPER) shipped 2026-09-08 with LOCAL_WHISPER as primary" ## Source code expectation `app/src/main/java/com/voicetranscriber/service/SpeechBackend.kt:62-67` defines the default `priorityOrder` as 4 entries: ```kotlin data class BackendConfig( val priorityOrder: List<SpeechBackendType> = listOf( SpeechBackendType.ANDROID, SpeechBackendType.FIREOS, SpeechBackendType.CLOUD, SpeechBackendType.LOCAL_WHISPER ), ``` `BackendConfig.load()` at line 138-162 returns either the saved order or the default. After `pm clear`, the prefs are empty so `order` is empty and `load()` returns `BackendConfig(settings = settings)` — which uses the data class default of all 4. The screen should render 4 cards. ## Models screen proves the type exists The Models screen (`Manage → Models`) does correctly list Whisper models for download (Tiny, Tiny (English), Base) and renders an "Active model: No model installed" empty state. The LOCAL_WHISPER backend type is fully wired in code: - `SpeechBackend.kt:20` — `LOCAL_WHISPER("local_whisper")` enum entry - `TranscriptionService.kt:343` — handler - `BackendConfigScreen.kt:546` — display name "Local Whisper" - `BackendConfigScreen.kt:553` — description "On-device whisper.cpp model" But none of these reach the priority list UI. ## Impact - Users on FireOS cannot promote LOCAL_WHISPER to the top of the priority list, so the SpeechRecognizer failure on FireOS (see #143) cannot be worked around - The BackendConfigScreen shows only 3 backends even though the source says 4 — code/build drift or runtime filtering bug - The "primary" backend per AGENTS.md is unreachable through the UI ## Suspected cause Either (a) bytecode in the installed APK differs from the source tree (the source HAS 4 backends but the compiled APK has 3), or (b) the `itemsIndexed(priorityOrder, key = { _, backend -> backend })` in BackendConfigScreen.kt:194 is dropping LOCAL_WHISPER at runtime, or (c) the encrypted prefs on this device had a saved order from a prior install that excluded LOCAL_WHISPER — but this was not reproducible after `pm clear`, so (c) is ruled out. ## Suggested fix 1. Compare the installed APK bytecode against the source tree to determine which of the 3 backends is missing and why. 2. Add a regression test that asserts `BackendConfig.default.priorityOrder.size == 4` AND that the screen renders 4 cards. 3. After the fix, ensure that LOCAL_WHISPER is the default FIRST backend in the priority order on FireOS (since the Alexa voice service binding fails on Fire devices — see #143). Reported by Hermes on 2026-09-10 during smoke testing of v1.0 (build from 93b7fd1) on Fire HD KFTRWI.

Mark added the

bug

label

2026-09-11 04:08:16 +00:00

Mark added the

triage-approved

label

2026-09-14 14:04:18 +00:00

Mark referenced this issue from a commit

2026-09-14 17:08:11 +00:00

fix(service): render all 4 backend cards on BackendConfigScreen (closes #144)

Mark closed this issue

2026-09-14 17:08:11 +00:00

Mark referenced this issue from a commit

2026-09-14 17:08:11 +00:00

Apply changes from kilo-gsd.py run for Mark/transcriber#144

Mark commented

2026-09-14 17:08:49 +00:00

Author

Owner

Fix landed in f2d2ff6 (fix(service): render all 4 backend cards on BackendConfigScreen (closes #144)), merged via d34cbf3.

Push recovered by watcher after kilo-gsd exited PARTIAL (non-fast-forward against remote tip 6e8e6a4 from #143). Merge of origin/main into local carried the #144 fix forward; push completed.

Refs: #143

Fix landed in f2d2ff6 (`fix(service): render all 4 backend cards on BackendConfigScreen (closes #144)`), merged via d34cbf3. Push recovered by watcher after kilo-gsd exited PARTIAL (non-fast-forward against remote tip 6e8e6a4 from #143). Merge of origin/main into local carried the #144 fix forward; push completed. Refs: #143