Skip to content

Latest commit

 

History

History
43 lines (34 loc) · 1.12 KB

File metadata and controls

43 lines (34 loc) · 1.12 KB
type canonical
source none
sync none
sla none
authority canonical
audience
contributors
maintainers
agents
last-verified 2026-04-16

SSOT

Current state

  • fallax is a Python-first reasoning-evaluation repo centered on the fallax package.
  • 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.

Active boundaries

  • fallax/ is the canonical evaluation-engine surface.
  • benchmarks/ holds reproducibility and regression material.
  • dashboard/ and website/ are presentation layers, not the scoring source of truth.
  • docs/ carries architecture, deployment, roadmap, and troubleshooting notes.

Active decisions

  • 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.

Validation baseline

pytest tests/
ruff check .
mypy fallax