Ollama fallback to cloud embeddings fires 472 Sentry errors — should be info/warn

[graycyrus]

Summary

When ollama embeddings are opted-in but the daemon is unreachable, the code correctly falls back to cloud embeddings. But it fires a Sentry error on every fallback — 472 events. A successful fallback is not an error.

Problem

What happened: User enables local ollama embeddings, but ollama isn't running. The code detects this and falls back to cloud embeddings (correct behavior). But each fallback fires a Sentry error:

ollama embeddings opted-in but daemon unreachable at http://localhost:11434; falling back to cloud embeddings for this session

Expected: This is an expected degraded state, not an error. Should be logged at INFO or WARN level, not ERROR. Sentry should not be flooded.

Impact: 472 events (TAURI-RUST-B), ongoing across 0.54.0 → 0.56.0.

Acceptance criteria

• Ollama fallback doesn't fire Sentry errors
• Unit test
• Diff coverage >= 80%

Related

• Sentry: https://sentry.tinyhumans.ai/organizations/tinyhumans/issues/65/

[oxoxDev]

Verified fixed on main. The ollama→cloud embedding fallback no longer fires a Sentry error per occurrence.

Path on main:

• memory_store/factories.rs:52-66 routes the daemon-unreachable report through observability::report_error_or_expected (not the raw capture that bypassed the classifier), with a process-scoped one-shot guard.
• observability.rs:880 GX arm matches ollama embeddings opted-in but daemon unreachable at → ExpectedErrorKind::ProviderUserState → info! only, no sentry::capture_message.
• Tight matcher: real ollama 500s (ollama_500_wire_shape_stays_unexpected) still reach Sentry; only the opted-in-but-down case is demoted.

Fixed by #2923 (and preserved through the #2959 suppression-revert, which only reworded the path here). RCA bucket: genuinely-unpreventable user state — user opted into local ollama but the daemon isn't running, zero local lever; the cloud fallback is correct, oscillation-guarded behavior, not a defect mask. Closing.