Reusable ledgers improve the efficiency of Agent trajectory policy auditing
Start here
In October 2026, an arXiv paper proposes a "write once, check many times" framework: a writer model converts each Agent trajectory into a policy-agnostic ledger, and a reader model checks a new policy using only the ledger, without re-reading the original trajectory. The authors test on 900 trajectories from SHADE-Arena, τ-bench, and τ²-bench; a 12B reader model reading narrative ledgers achieves an AUC that matches or exceeds that of the same model reading full trajectories; on SHADE-Arena policies requiring combining multiple facts, the method achieves an AUC of .777, exceeding a 31B model reading full trajectories.
Generated by AI from reports · Updated 24 days ago
Attention and developments
Attention, hourly
Now 33, 0% in 24 hours·Peak 33 (Nov 7 01:00)
Related research
This event's paper
- Write Once, Check Many: Reusable Ledgers for Post-Hoc Policy Compliance Auditing of Agent Trajectories
arXiv · 2026-10-17 · Deep reading
Progress timeline
- arXivWrite Once, Check Many Times: A Reusable Ledger for Post-Hoc Policy Compliance Auditing of Agent Trajectories
The paper proposes a "write once, check many times" framework, in which a writer model converts each Agent trajectory into a policy-agnostic ledger, and a reader model checks a new policy using only the ledger, without re-reading the original trajectory. The authors test on 900 trajectories from SHADE-Arena, τ-bench, and τ²-bench: when a 12B reader model reads narrative ledgers, its AUC matches or even exceeds that of the same model reading full trajectories; on SHADE-Arena policies that require combining multiple facts, the method achieves an AUC of .777, exceeding the .713 of a 31B model reading full trajectories. When checking 8 SHADE-Arena policies individually, the ledger still uses about 5 times fewer tokens even counting the writing cost. The analysis also finds that rule-based extraction methods perform better on some single-action policies, and that ledgers may miss evidence in longer tool outputs and message content.