python 3.12.15 numpy 2.5.4 rank_bm25 0.2.2
chunks=1504 avg_len=137.9 tokens, vocab=9838
queries: 300 heading (2+ tokens), 200 long (8 tokens)

[correctness, robertson idf, 500 queries x 1504 docs]
max abs score difference: 1.776e-13
top-10 set identical: 500/500; top-1 identical: 500/500
of 0 queries with a different top-10 set, 0 differ only by a score tie at rank 10

[lucene idf vs rank_bm25] mean top-10 overlap 0.895; top-1 identical 486/500

[speed: ms per query, best of 3 passes]
| query set | rank_bm25 | from scratch (inverted index) | speedup |
|---|---:|---:|---:|
| heading | 1.405 | 0.810 | 1.7x |
| long | 1.662 | 1.436 | 1.2x |

index build: rank_bm25 39 ms, from scratch 58 ms