All configurations / harness
BM25 top-20 + rerank
- Configuration
- A frozen BM25 top-20 pool reranked by the model to exactly five IDs with the baseline prompt.
- Source
- campaign/verify_find.py
- Recorded runs
- 7 scored outputs · $0.20 measured model spend · runs on 2026-09-15
- Access
- open harness · standard tools
FIG. 01
By dimension
Scores /10 · Paired p-values are exploratory and unadjusted—Verify
Not yet tested.
—Ground
Not yet tested.
4.38Find
Baselinen=4 · recall@5 · $0.0302/task · 3.1s/task · 2026-09-15
headline; 4 tasks, 4 source clusters; 0 failed attempts kept in the denominator; free BM25 0.354; candidate oracle 0.500; 2 of 4 pools reachable
Run provenance
Report path: outputs/campaign/20260915/summary.json
Task pack: campaign-dev-20260915
Run: campaign-v3
- targeted recoverable misses10.00 · n=2 · $0.0251 · 2026-09-15 · targeted recoverable misses; 2 tasks, 2 source clusters; 0 failed attempts kept in the denominator; free BM25 0.000; candidate oracle 1.000; 2 of 2 pools reachable
- retrieval-ceiling control10.00 · n=1 · $0.0294 · 2026-09-15 · retrieval-ceiling control; 1 tasks, 1 source clusters; 0 failed attempts kept in the denominator; free BM25 1.000; candidate oracle 1.000; 1 of 1 pools reachable
—Extract
Not yet tested.
—Analyze
Not yet tested.
—Compute
Not yet tested.
—Synthesize
Not yet tested.
—Review
Not yet tested.
—Hypothesize
Not yet tested.