Repository navigation
1.5.1 — Quota forecast recovery and pricing diagnostics
- Restored quota projections when credit calibration is blocked by unpriced calls, using time-weighted measured quota changes as a fallback. The projection explicitly reports percentage-per-hour/day rates rather than inventing credit amounts.
- Added safeguards for observations older than 24 hours, insufficient coverage, quota decreases, long gaps, and inconsistent current values. Observed history remains separate from the projection.
- Added grouped pricing diagnostics (reason, model, service tier, count and date range) to the pricing dialog and calculation export; exports from the quota page now use its selected quota period.
- Kept unpriced calls excluded from monetary and credit totals. This release restores eligible forecasts without requiring source-log reinjection; it does not assume that every unpriced call has a known rate.
Docker: capitaine/codex-usage-dashboard:1.5.1 and :latest share the same AMD64/ARM64 image index, sha256:e86bf6879bc0e2860716669852a65f285be6d915fbcb31d47ed956661bbfdc41.
Pull the image and recreate the hub container while preserving configuration and storage. The forecast fix uses quota observations already received by the hub; no source-log reinjection is required when that history is sufficient.
Validation: 258 root tests, Linux/Windows CI, desktop/mobile browser checks, and a hardened published-image Mesh hub smoke test including the 56%/869-unpriced-calls regression fixture passed. The fixture is synthetic; the original user's unpriced-call reason remains to be identified with the new diagnostics.