Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Trace2Policy: From Expert Behavior Traces to Self-Evolving Decision Agents

About

Decision rules that enterprise experts apply tacitly -- in auditing, compliance, and contract review -- can be systematically recovered and improved through iterative error analysis. We present \textbf{Trace2Policy}, whose core mechanism -- \textbf{EISR} (\textbf{E}rror-driven \textbf{I}terative \textbf{S}kill \textbf{R}efinement) -- maintains a human-readable rule document as its optimization target: each round executes the rules on a validation set, clusters errors by root cause into MISSING, WRONG, or CONFLICT types, applies targeted patches, and commits only those that pass a regression gate. \textbf{For this class of compliance-sensitive, skewed-base-rate decision tasks, we identify rule quality -- not model capability -- as the dominant performance lever}: across five LLMs, one-shot distillation plateaus near $\sim$70\% on the deployed pool, while eight EISR rounds lift the same rules to 79.6\% when compiled into deterministic Python -- zero LLM calls at inference. \textbf{Execution form compounds the gain: in production, the same EISR-refined content runs 9.8~pp higher as compiled Python than as an LLM prompt, a form-and-engineering bundle the 22-day deployment matured together.} Deployed for 22 days at a major logistics carrier (3,349 audit cases), the compiled pipeline outperforms the pure-LLM baseline it replaced (72.7\%); on these calibrated, skewed-base-rate workloads, re-enabling LLM fallback monotonically degrades accuracy. An LLM-driven variant, \textbf{Auto-EISR}, reproduces this refinement at \$5--\$10 per cycle versus $\sim$70 expert-hours, and transfers to four public benchmarks spanning legal reasoning (LegalBench) and process-mining decisions (BPIC 2012) without re-engineering.

Junli Zha, Jinbo Wang, Chao Zhou, Xiang Song• 2026

Related benchmarks

TaskDatasetResultRank
Decision classificationDamage-audit 139-case extended (held-out)
Action Accuracy81.3
8
Audit-relevance Classificationbenchmark_1000 n=993 (train)
Accuracy79.6
5
decision predictionBPIC loan-decision 2012
Accuracy84.7
3
Natural Language Inferencecontract_nli LegalBench
Accuracy81.7
3
Decision classificationDamage-audit (40-case held-out)
Action Accuracy76.4
3
hearsay classificationLegalBench Hearsay
Accuracy72.7
3
unfair terms detectionLegalBench unfair_tos
Accuracy81.4
3
Audit-relevance Classificationnew n=1462 2000 (held-out)
Accuracy77.3
2
Audit-relevance Classificationrecent_week n=894 (drift)
Accuracy76.8
2
Showing 9 of 9 rows

Other info

Follow for update