| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| director_ai-3.9.5-py3-none-any.whl | 2026-03-23 | 396.7 kB | |
| director_ai-3.9.5.tar.gz | 2026-03-23 | 581.0 kB | |
| sbom.json | 2026-03-23 | 72.8 kB | |
| README.md | 2026-03-23 | 998 Bytes | |
| v3.9.5 source code.tar.gz | 2026-03-23 | 10.9 MB | |
| v3.9.5 source code.zip | 2026-03-23 | 11.4 MB | |
| Totals: 6 Items | 23.3 MB | 0 | |
What's Changed
Features
- Coalesced
review_batch(): batches logical and factual NLI pairs throughNLIScorer.score_batch()— 2 GPU forward passes total instead of 2*N. Dialogue items fall back to sequentialreview(). - Per-task-type judge escalation:
_should_escalate()now receives the detected task type at all 4 call sites. Per-task thresholds: dialogue=0.35, summarization=0.25, qa=0.30, fact_check=0.20. - 17 new tests (no-KB calibration, cross-turn blending, judge threshold differentiation).
Fixes
- Scorer:
_finalise_review()log message uses effective threshold override instead of base threshold. - Scorer: meta-classifier NLI-to-coherence threshold conversion uses actual weights instead of hardcoded 0.4/0.6.
- 50+ documentation corrections across docs-site, notebooks, README, ROADMAP, CONTRIBUTING, CITATION.cff — verified by dual audit (Claude + Gemini).
Full Changelog: https://github.com/anulum/director-ai/compare/v3.9.4...v3.9.5