Contract analysis
ContractEval
Clause-level legal-risk identification over 41 risk categories, including a laziness metric for answering that no related clause exists when one does.
As of 13 August 2026. Conditional pack. ContractEval methodology on CUAD. Laziness, correctness, and output effectiveness stay separate. Inherit CUAD terms.
ContractEval scores clause-level legal-risk identification over 41 risk categories. Laziness is the unusual metric: answering that no related clause exists when one does. That is a different failure from a wrong span.
The pack is a paper and harness on CUAD, not a new corpus. CUAD license and contamination apply. No Fable 5 or GPT-5.6 table is cited here.
- Scale
- Methodology and results paper over CUAD.
- Provenance
- olivialiu121/ContractEval. Runs on CUAD; inherit CUAD terms.
- Contamination
- High. Built on public CUAD.
- Cannot establish
- Legal correctness in a client fact pattern, or Harvey product performance. Not a separate corpus from CUAD.
Metrics, kept separate
- Correctness
- Output effectiveness
- Laziness
No cited public figure is installed for this pack. Method, scale, and gates still apply. A missing chart is not a pass rate.
What the figures bound
Use this pack, after rights clearance, to keep refusal-to-work separate from extraction error. Do not treat it as Harvey, and do not average laziness into CUAD AUPR.
No chart is installed without a cited official cell from the paper replayed on a frozen revision.
What this page does not conclude
- Legal correctness in a client fact pattern
- Harvey or CLM product performance
- GPT-5.6 or Fable 5 laziness on CUAD
- A single ContractEval winner from mixed metrics