LegalBench
Fits contract-review, compliance and legal-research assistants that must classify clauses or apply a rule to facts. Borrow its six-way reasoning taxonomy (issue, rule, conclusion, application, interpretation, rhetoric) and its short hand-written task cards.
What it asks the agent to do
- spot the legal issue in a short fact pattern
- recall a legal rule and apply it to facts
- classify a contract clause or privacy-policy statement
- decide whether a clause creates a given obligation
- interpret a statutory or contractual text
What to borrow from it
taxonomycase templatebaseline
Caveats
Question-answer only, no tools or multi-step work; 162 tasks are mostly US-English classification with per-task licenses inherited from upstream datasets (CUAD, MAUD, CaseHOLD etc.), so license is mixed; top models score near 89 percent on the Vals leaderboard, so it is close to saturated.
Details
| Workloads | legal, knowledge qa, documents |
|---|---|
| Interaction | single turn |
| Environment | static |
| End state checked | No |
| Tools | none |
| Grading | exact match, programmatic check |
| Size | 162 |
| Data | mixed |
| Personal data | none |
| License | unknown |
| Harness | own repo |
| Status | active |