| type | canonical | |||
|---|---|---|---|---|
| source | none | |||
| sync | none | |||
| sla | none | |||
| authority | canonical | |||
| audience |
|
|||
| last-verified | 2026-04-16 |
fallaxis a Python-first reasoning-evaluation repo centered on thefallaxpackage.- The runtime surface includes the evaluation engine, benchmark suites, a dashboard, and a small project site.
- Docs and repo surfaces are now aligned with the benchmark and scoring semantics described in the current README and docs lane.
fallax/is the canonical evaluation-engine surface.benchmarks/holds reproducibility and regression material.dashboard/andwebsite/are presentation layers, not the scoring source of truth.docs/carries architecture, deployment, roadmap, and troubleshooting notes.
- Step-level correctness matters more than single final-answer scoring.
- Benchmark semantics must stay explicit and reproducible.
- New domains or task packs should land with docs and tests, not as silent config drift.
pytest tests/
ruff check .
mypy fallax