Skip the engine warmup when the pinned recorder will record

[r3dbars]

Why

#1857 moved built-in and wired mics onto the pinned recorder. But pinnedDictationSkipsEngineWarmup() still only skipped the engine when the macOS input was a Bluetooth headset.

So after a wake or route change (readiness stale), a key press warmed AVAudioEngine on the default mic, waited for it, then recorded through the pinned recorder anyway. That start never uses the engine.

From local events.jsonl on 1.1.65+:

• Pinned starts: p50 111 ms / p95 138 ms (120 starts).
• The 2 starts that hit the stale wait: 233 and 327 ms. debug.log shows started after 289ms wait while the engine warmed the MacBook mic, then recording started on the pinned microphone.

What changed

• One rule for warmup and recording. PinnedDictationInputPolicy.skipsEngineWarmup(for:afterEngineFallback:) decides warmup from the same pinnedDictationInputSelection() and recorderIsNeeded rule the start uses, so the two can't drift apart again.
• Bluetooth guard kept. A Bluetooth macOS input (or an unreadable route) still never warms the engine, so idle AirPods never get pulled into call mode.
• Warm again after a fallback. When a pinned start falls back to the engine, pinnedDictationFellBackToEngine turns warmup back on until the recorder next starts. That way a repeat fallback isn't also a cold start.
• Updated the stale comments and the line in Sources/Speech/CLAUDE.md.

Tradeoff: built-in and wired mic users no longer get the engine prewarmed at launch or wake. So the first fallback after launch starts cold. Field data: 4 fallbacks in 117 pinned starts over 30 days, all from one install. Local: 0.

Verification

• bash build.sh --no-open: pass
• bash run-tests.sh: 18514 / 18514 pass
• python3 scripts/dev/check-source-pins.py --changed-only: pass
• Mutation check: removing the fallback clause fails the new test, so the test catches it.

Not verified on hardware. A live A/B (switch the macOS input to the MacBook mic, then dictate) was started but not finished. Plan: after merge, compare pinned starts that follow a route change, plus dictation_started_after_wait counts, in events.jsonl before and after.

🤖 Generated with Claude Code