## Summary
- The 15-minute session keep-alive interval in `useDraftAuthRecovery` had no `try-catch`. When `authClient.getSession()` is called while the network is down (e.g. a laptop waking from sleep), `better-fetch` throws a `TypeError: Failed to fetch` — producing an unhandled promise rejection that was reported in Sentry.
- Fixed by wrapping the interval callback in a `try-catch`. Network-level `TypeError`s are silently swallowed because real session expiry (HTTP 401) is returned in-band as `{ data: null }` by `better-fetch` and already caught by the `!session` check.
- Improved the comment to explain the mechanism rather than citing socket offline detection (which doesn't validate auth).
- Added 3 tests covering: network throw (no authDegraded), null session response (authDegraded set), valid session (no change).
## Root cause
User left the draft page open overnight. The 15-minute ping fired repeatedly. At some point the computer woke from sleep with the network not yet ready — `getSession()` threw, `setInterval` dropped the rejected promise, Sentry caught it as an unhandled rejection. Confirmed via Sentry breadcrumbs: last activity May 26 8:32 PM, error fired May 27 12:23 PM (~28 hours later).
## Test plan
- [x] 3 new tests pass
- [x] typecheck clean
- [ ] After deploy: monitor Sentry for recurrence
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Chris Parsons <chrisparsons1127@gmail.com>
Reviewed-on: #56
chrisp
8a40953019