Numeric exactness
100%
Gate: 100% · min 100% · PASS
60-day paid pilot · measuredAll hard gates pass
Gold-set evaluation results
Generated Jul 16, 2026, 4:06 PM from 10 gold-set cases. These figures come from python -m anvil_worker.eval_runner, not an illustrative example.
Critical finding recall
100%
Gate: 100% · min 100% · PASS
Valid evidence rate
100%
Gate: ≥98% · min 100% · PASS
Unsupported approved findings
0
None found across the suite
This report measures extraction accuracy against the gold-set corpus only. Time reduction is not computed by this harness — it must be measured operationally during the pilot by timing manual versus assisted review of the same bid packages. Material-scope recall (mean 98%, min 83%) is tracked in the suite but is not one of the three hard release gates.
Per-case results
Every gold-set case in this run, scored independently
| Case ID | Document | Numeric exactness | Critical recall | Material recall | Evidence validity | Unsupported approved |
|---|---|---|---|---|---|---|
| acme-plumbing-underground-01 | evals/gold/case_01_plumbing.pdf | 100% | 100% | 100% | 100% | 0 |
| northside-drywall-level4-02 | evals/gold/case_02_drywall.txt | 100% | 100% | 100% | 100% | 0 |
| volt-electric-alternates-03 | evals/gold/case_03_electrical.txt | 100% | 100% | 100% | 100% | 0 |
| summit-mechanical-ductwork-04 | evals/gold/case_04_mechanical.txt | 100% | 100% | 100% | 100% | 0 |
| bedrock-concrete-commercial-05 | evals/gold/case_05_concrete.txt | 100% | 100% | 83% | 100% | 0 |
| ironclad-steel-messy-06 | evals/gold/case_06_steel_messy.txt | 100% | 100% | 100% | 100% | 0 |
| glenwood-addendum-conflict-07 | evals/gold/case_07_addendum.txt | 100% | 100% | 100% | 100% | 0 |
| delta-glazing-ambiguous-total-08 | evals/gold/case_08_glazing_typo.txt | 100% | 100% | 100% | 100% | 0 |
| apex-fire-sprinkler-letter-09 | evals/gold/case_09_fire_sprinkler.pdf | 100% | 100% | 100% | 100% | 0 |
| garbage-near-empty-10 | evals/gold/case_10_garbage.txt | 100% | 100% | 100% | 100% | 0 |