I'd like to test 2 configs:
- 3+ terms with 80% on stems and 66% on plain 80% on stems, expected ZRR change: -4.6% (1k sample enwiki search queries)
- 4+ terms with 80% on stems and 66% on plain 80% on stems, expected ZRR change: -2.1% (1k sample enwiki search queries)