OpenEval v0.1.3 makes large local transcript collections faster to inspect, easier to operate, and much harder to misread.
Performance
- 93.5% smaller Live list payload: the measured 100-session response fell from 2.65 MB to 171 KB. Full trace, tool, queue, file, and usage detail now loads only when a session drawer opens.
- 57% faster known-source discovery: same-machine median improved from 60 ms to 26 ms.
- Exact Gemini inventory: 707 broadly matched JSON files became 1 verified
~/.gemini/tmp/**/logs.jsonartifact. - Unknown-source discovery is reused for 30 seconds while known-source fingerprints and content sentinels remain fresh.
Data fidelity and proof
- Collection separates parseable and detect-only files and reports measured, inferred, missing, malformed, stale, archived, and incomplete evidence.
- Parse budgets, parser safety caps, and bounded detect-only walks visibly mark partial data instead of silently looking complete.
- Live reports the exact discovered, scanned, parsed, dropped, and unscanned population behind its current slice.
- Timeline distinguishes signal, judged, heuristic, and no-signal sessions and shows the actual outcome denominator used by every median.
- Artifact responses include byte count, SHA-256, modification time, ETag, and 304 support without conflating byte integrity with visual quality.
- Visual coverage, expected-artifact contracts, evidence tiers, and strict accuracy gates now fail closed against stale or structurally weak proof.
UX and accessibility
- Resilient loading, stale, retry, and error states across Dashboard, Collection, Live, Timeline, and Accuracy.
- Accessible case rows, checkbox/re-run controls, focus-trapped session drawers, focus restoration, body-scroll locking, and keyboard navigation.
- The sidebar version now comes from package metadata and displays
v0.1.3. - README and changelog now document the release's measured improvements and evidence boundaries.
Security
- Live drawer and transcript requests resolve client-returned paths through the current server-owned source inventory before parsing.
- Out-of-source and symlink-escape paths are rejected, closing the two high-severity CodeQL path-flow alerts raised during release review.
Verification
- Full
npm test - TypeScript typecheck and zero-warning lint
- Strict accuracy: 24/24 oracle, 24/24 known-bad, 24/24 deterministic/trace coverage
- Self-test: 49 pass, 0 fail, 6 expected opt-in LLM-judge skips
- Production build and public-upload audit
- Chrome runtime proof of the
v0.1.3sidebar, with no console errors or horizontal overflow - GitHub CI and all CodeQL checks passed before merge
Full changelog: v0.1.2...v0.1.3