Unassigned · 2026-10-09
Safe, Persistent, and Evolving Agent Harness for Understanding Partially Observable Worlds
Yisen Gao, Yue Guo, Qing Zong, Yiwen Guo, Yangqiu Song
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
The authors introduce E-LEDGER, a multi-agent harness for safe, persistent execution in partially observable enterprise workflows. It compiles policies into executable code approvals, and maintains a world ledger with verified hidden rules plus evidence-backed dynamic state. They further propose WorldAbduct, which uses four diagnostic views and targeted abduction interactions to recover hidden rules for the ledger. On WOW, E-LEDGER+WorldAbduct improves STCR by outperforming the strongest evolution baseline by 5–15 percentage points.
Why it matters
Enterprise tool-using agents can take actions that are locally plausible but disallowed by policy, and tool feedback may hide side effects or delayed changes under partial observability. The proposed harness makes policy enforcement non-bypassable via code approval and maintains persistent state plus verified hidden rules. WorldAbduct addresses the fact that hidden rules are unknown initially by diagnosing execution gaps and validating hypotheses through targeted environment interactions before adding them to the ledger, aiming to improve safety while completing long-horizon tasks.
Method
- E-LEDGER: multi-agent harness with a world ledger (static hidden rules + dynamic evidence-backed state) and a code-based approval layer that checks every proposed action against compiled organizational policies before execution.
- WorldAbduct: abductive harness evolution that diagnoses execution trajectories using four views (state consistency, world-observation gap, policy-gate correctness, goal judgment), then verifies hidden-rule hypotheses via targeted abductive interactions before adding them to the static world layer.
- State tracking updates dynamic ledger from tool responses using applicable static rules, recording conflicts/unresolved fields for later verification; approval outcomes can be execute/query/probe/block to drive repairs or evidence gathering.
Limitation
The interactive environments make full evolution and evaluation costly, so the authors report single-run results for each configuration.
Abstract (from arXiv)
Large language model agents can invoke tools fluently, but enterprise workflows demand more than selecting the right tools: actions must strictly comply with organizational policies, tool feedback often conceals hidden side effects under partial observability, and long-horizon tasks require persistent state tracking across multiple records. To address these challenges, we introduce E-Ledger, a multi-agent harness for safe and persistent execution. E-Ledger employs a code approval layer that checks every proposed action against policy before execution, and maintains a world ledger of verified hidden rules alongside evidence-backed dynamic state. Because hidden rules are typically unknown a priori, we further propose WorldAbduct, an abductive, world-model-driven harness evolution framework. WorldAbduct diagnoses execution trajectories across four complementary views (state consistency, world-observation gap, policy-gate correctness, and goal judgment) to hypothesize latent rules, and verifies them through targeted abductive interactions before integrating them into the ledger. On the enterprise benchmark World of Workflows, E-Ledger with WorldAbduct improves safe task completion across four LLM backbones, outperforming the strongest evolution baseline by 5--15 percentage points. Experiments in ScienceWorld and DiscoveryWorld further show that abductive harness evolution carries over to scientific environments. Our code is available at https://github.com/HKUST-KnowComp/E-LEDGER-WorldAbduct.
Related papers
- ANTE: Anticipate, Then Execute -- Learning Internal World Models for Coding Agents from Verified Beliefs
- Failure Recovery via Belief Updates and World Modeling for Long-Horizon Agents
- Agent World Model in the Wild: Computer-Use Agent Environment Synthesis
- EvoBelief: Falsifiable World Models for Self-Improving LLM Agents
- Kepler: Auditable World Models for ARC-AGI-3
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-08 · Indexed by Pub-AI: 2026-10-09 UTC
Safe, Persistent, and Evolving Agent Harness for Understanding Partially Observable Worlds
Preprint posted: 2026-10-08 · Original record ↗
- arxiv
- 2610.11552
Explicit source versions (1)
- v1 · 2026-10-08
- Preprint available
Source date: 2026-10-08 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Large language model agents can invoke tools fluently, but enterprise workflows demand more than selecting the right tools: actions must strictly comply with organizational policies, tool feedback often conceals hidden side effects under partial observability, and long-horizon tasks require persistent state tracking across multiple records. To address these challenges, we introduce E-Ledger, a multi-agent harness for safe and persistent execution. E-Ledger employs a code approval layer that checks every proposed action against policy before execution, and maintains a world ledger of verified hidden rules alongside evidence-backed dynamic state. Because hidden rules are typically unknown a priori, we further propose WorldAbduct, an abductive, world-model-driven harness evolution framework. WorldAbduct diagnoses execution trajectories across four complementary views (state consistency, world-observation gap, policy-gate correctness, and goal judgment) to hypothesize latent rules, and verifies them through targeted abductive interactions before integrating them into the ledger. On the enterprise benchmark World of Workflows, E-Ledger with WorldAbduct improves safe task completion across four LLM backbones, outperforming the strongest evolution baseline by 5--15 percentage points. Experiments in ScienceWorld and DiscoveryWorld further show that abductive harness evolution carries over to scientific environments. Our code is available at https://github.com/HKUST-KnowComp/E-LEDGER-WorldAbduct.