@microsoft-github-policy-service agree company="0din"

Neat work!

Thanks for putting this together! Two suggestions on the scorers side (the seed datasets look great):

1. Subclass RegexScorer instead of adding _JEFKeywordScorer. The new base class is functionally equivalent to RegexScorer with re.escaped patterns — same OR-join, same true_false output. Subclassing RegexScorer would shrink each scorer to ~10 lines and reuse its "Matched: <name1>, <name2>" rationale, which is more useful for triage than the generic hit/miss strings. We''d just pass patterns={kw: re.escape(kw) for kw in METH_KEYWORDS} (or pre-escape the dict at module scope) and categories=["meth"].
2. Drop the JEF prefix from the class names. MethKeywordScorer, NerveAgentKeywordScorer, etc. — JEF isn''t a widely-known acronym and users will mostly find these by the substance name. We''d absolutely keep the attribution: module docstring, class docstrings, and See Also links pointing to https://github.com/0din-ai/0din-JEF stay as-is, and the keyword lists themselves remain credited to JEF.

Would you like to make these changes, or would you prefer we take it from here? Happy either way — if we don''t hear back, we''ll pick it up sometime in June.

Thanks for the suggestions! Both addressed:

1. Scorers now subclass RegexScorer — each is ~10 lines with {kw: re.escape(kw) ...} patterns and substance-specific categories.
2. Dropped JEF prefix from class names and split into individual files (meth_keyword_scorer.py, etc.) to match codebase convention. Attribution stays in module and class docstrings.

Also rebased onto current main to pick up RegexScorer from #1704.