Investigate forecast calibration read deadlines on v0.155.1-beta.1 #1830

On v0.155.1-beta.1 (Core commit 16c4ad090d3ad7a4735f473d4e0991b97593dd43), two new forecast calibration unavailable warnings reported context deadline exceeded on 25 October 2026. Their journal timestamps are 08:00:41.495 and 08:15:38.953 UTC.

In that exact source, forecastTracker.refreshEvidence in go/cmd/ftw/forecast_tracking.go:399 logs a failed LoadForecastErrors call and returns before loading observations or replacing the cached evidence. The function sets a three-second read budget. This is a source-code budget, not a measured SQL or lock duration.

The exact SQL step, lock owner, IO/CPU cause and recovery of each call remain unknown. No numeric SQLite error code was established. The later healthy status does not prove that either calibration read recovered. Writer and maintenance failure counters did not increase during the observed period. These warnings do not establish the cause of the older history write failures in #1554 or of the token persistence failure addressed by #1826.

Please identify the failing read stage with bounded diagnostics, reproduce it with a small anonymous test database, and test timeout handling plus recovery/freshness of the cached calibration. Keep private site data and raw database contents out of the issue. This report contains observations and a source check; it is not a tested fix or hardware acceptance.