brrrn v0.1.2
The numbers got audited. This release is the result of an end-to-end audit of brrrn's cost and token accounting (headless Codex at x-high reasoning, then adversarially reviewed commit by commit before merge). Every fix carries a test that fails without it, and the corrected totals were reconciled against real logs.
On the audited corpus, brrrn had been undercounting by about $2,028 all-time. If your numbers jump after updating: they were always yours, we just weren't counting them all.
Fixed
- ⚖️ Priority/flex tiers priced correctly: Codex priority processing now uses LiteLLM's tier-specific rates (about 2x standard) instead of standard rates. This was the bulk of the undercount.
- 🧾 Claude fallback/compaction iterations are billed per iteration; free pre-output refusals excluded; finalized usage beats partial streaming copies from resumed sessions.
- 🔥 Streak math is stable at microdollar precision everywhere: fifty $0.10 entries can no longer sum to $4.999999... and break your streak.
- 🧮 Hub rounding preserves decimal half-ties; app model-merge keys can no longer collide.
Faster
- App refresh: one engine scan instead of four (warm 0.22s → 0.13s, cold CPU 26.3s → 6.0s).
Notes
- Cache format bumped to v7: the first scan after updating does a one-time full rescan (~8s on large histories), then returns to ~0.1s.
- Unsigned bundle: first launch needs right-click → Open.
shasum -a 256 BrrrnBar.app.zip
b63f15c95565370e9ec6ba5dd212470be037d1d6c19d3ae35a3d44c81c2255d1
Full details in the CHANGELOG.