Skip to content

v0.2.0

Choose a tag to compare

View all tags
@adaamko adaamko released this 01 Jul 20:27
· 15 commits to main since this release
v0.2.0
888f275
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Highlights

Learning that generalizes

  • Dev-holdout refinement: patches are accepted only if they improve quality on a held-out split (learn_rules(holdout_fraction=0.2)), with stratified splitting and failure-mode-clustered patch prompts.
  • Per-rule validated precision & support, Wilson-bounded ranking (rank_rules()), and precision-based conflict resolution in the executor.

Robustness

  • Catch-all regex ban, truncated-JSON rule recovery, and per-rule execution timeouts — an LLM-written rule can no longer hang the loop.
  • regex is now a core dependency (matching timeout guard).

Load & inspect rulesets

  • Rule.from_dict and RuleChef.load_rules(...): load a saved ruleset from a file/dict/list and immediately extract or evaluate with it.
  • benchmarks/rule_report.py: browsable HTML report of a ruleset against gold data — per rule TP/FP with the matched span highlighted in context. See benchmarks/INSPECTING_RULES.md.

Evaluation

  • Benchmark harnesses + committed results in benchmarks/results/ (TAB anonymization, ablation, HITL feedback repair, observation mode, and more).

Contributor infrastructure

  • Issue forms with native types, PR rights-attestation checkbox enforced by CI, advisory DCO sign-off warning, Code of Conduct.

Install: pip install rulechef==0.2.0 · PyPI: https://pypi.org/project/rulechef/0.2.0/

Assets 3
Loading