Agents & procedures

Benchmark first

Corpora: KILT, QASC, and LongMemEval, plus repository-owned golden tool traces for workflow replay.

Protocol: Separate retrieval/evidence quality from agent execution. Report task success, exact and partial tool-sequence match, argument validity, tool-error recovery, grounded-output rate, trajectory cost, and reused-procedure uplift. Evaluate mined procedures on held-out tasks and reject any train/test source overlap.

• Trajectories and agent evaluation
• Procedural learning and regression gates
• Procedural knowledge