Formal models and scripts: queries-mlt-commonterms-reference

Tools: Python

These are the working models and scripts this part of the hunt used. Some encode the code as it is (a counterexample is the expected result), others a fixed or correct variant; read the header comment of each file. Paths inside them were replaced by placeholders such as $LUCENE (a Lucene checkout) and $WORK.

Findings

• MoreLikeThis keeps the lowest-scoring selected terms when the query reaches the boolean clause limit, so the truncated query silently drops the most interesting terms (regression since 4.10.0) (finding LFV-571; repro · finding page)
• MoreLikeThis with setBoost(true) divides every term's score by the lowest selected score, so a selected term that scores 0 (a TFIDFSimilarity whose idf is 0, such as the textbook log(N/df) for a term in every document) makes like() throw 'boost must be a positive float, got NaN' (8.0 and later) or build NaN and Infinity boosts (4.10 to 7.x) (finding LFV-559; repro · finding page)
• MoreLikeThis with maxQueryTerms = 0 throws NullPointerException from like(...), retrieveInterestingTerms(...) and the rewrite of MoreLikeThisQuery or the XML parser's LikeThisQuery, instead of returning a query with no terms (regression since 4.10.0) (finding LFV-553; repro)

Files

• model/mlt_selection_model.py (used for LFV-553, LFV-559, LFV-571)

How to run

• Python: python3 <file>.py; scripts that import z3 need the z3-solver package.

Snapshot of 2026-11-03. AI-assisted; not yet triaged by a human Lucene committer. Feedback: open an issue in lucene-ai-bug-hunt.