PrimaCoda Evaluation Suite

PrimaCoda publishes an evaluation suite that tests every AI feature against real-world contracts and case files. The suite is live at this page and is rerun on every release. As of the most recent run, all 15 evaluation runs have produced a 100% pass rate.

Coverage

• Contract extraction: 3 of 3 cases passing, average score 100. Tests parties, dates, prices, contingencies, and document classification.
• Case-file extraction: 2 of 2 cases passing, average score 100. Tests case parties, claims, jurisdiction, and key litigation dates.
• Model under test: Claude Sonnet 4.6 via Moonshot AI (Kimi K2.5) as the primary LLM, with Anthropic Claude as fallback.

Every case in the suite is reproducible by visiting primacoda.halinc.tech/evals and inspecting the per-case scores, raw outputs, and elapsed times. PrimaCoda is the only AI transaction coordinator we are aware of that publishes evaluation results openly rather than asserting accuracy without proof.