ContractBench: Can LLM Agents Preserve Observation Contracts?

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Jicheng, He, Yifeng, Wang, Zili, Xing, Hanwen, De, Arkaprava, Chen, Hao
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916020390723584
author Wang, Jicheng
He, Yifeng
Wang, Zili
Xing, Hanwen
De, Arkaprava
Chen, Hao
author_facet Wang, Jicheng
He, Yifeng
Wang, Zili
Xing, Hanwen
De, Arkaprava
Chen, Hao
contents Tool-augmented LLM agents call APIs whose intermediate outputs, such as presigned URLs, session tokens, and OAuth state parameters, are observation contracts: artifacts whose later use is constrained by the external system that produced them. We show that observation contract compliance (preserving the temporal validity and byte-level integrity) is an emergent, regression-prone capability: it is neither guaranteed by general tool-use ability nor consistently improved by larger or newer models. To measure this, we introduce ContractBench, a benchmark of 33 dual-axis tasks that probe two orthogonal failure modes no existing benchmark evaluates: validity failures (using an artifact after expiry) and integrity failures (corrupting an artifact's bytes through the observation-to-action pipeline). Our evaluation is deterministic and programmatic, with a virtual clock controlling time and SHA-256 hashes verifying byte integrity. We assign each outcome a failure label drawn from real-world API specifications. We evaluate 38 models and report four findings: (i) no evaluated model clears 80%, with Claude-Opus-4.6 leading at 77.8%, revealing that current frontier models still fail to comply with observation contracts; (ii) a sharp within-family capability cliff in Qwen 3.5 between 4B (0%) and 9B (56.6%), smoothing to 70.7% at 397B-A17B: what emerges across the cliff is mid-trajectory restraint, not tool-call competence; (iii) non-monotonic scaling across the GPT-5 family: agentic post-training can erode compliance through sycophancy-driven regression; (iv) our failure taxonomy works as an actionable in-context reward signal, yielding +7.1 pp on 42 paired GPT-5.1 failures.
format Preprint
id arxiv_https___arxiv_org_abs_2605_17281
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ContractBench: Can LLM Agents Preserve Observation Contracts?
Wang, Jicheng
He, Yifeng
Wang, Zili
Xing, Hanwen
De, Arkaprava
Chen, Hao
Software Engineering
Artificial Intelligence
I.2.7; I.2.6
Tool-augmented LLM agents call APIs whose intermediate outputs, such as presigned URLs, session tokens, and OAuth state parameters, are observation contracts: artifacts whose later use is constrained by the external system that produced them. We show that observation contract compliance (preserving the temporal validity and byte-level integrity) is an emergent, regression-prone capability: it is neither guaranteed by general tool-use ability nor consistently improved by larger or newer models. To measure this, we introduce ContractBench, a benchmark of 33 dual-axis tasks that probe two orthogonal failure modes no existing benchmark evaluates: validity failures (using an artifact after expiry) and integrity failures (corrupting an artifact's bytes through the observation-to-action pipeline). Our evaluation is deterministic and programmatic, with a virtual clock controlling time and SHA-256 hashes verifying byte integrity. We assign each outcome a failure label drawn from real-world API specifications. We evaluate 38 models and report four findings: (i) no evaluated model clears 80%, with Claude-Opus-4.6 leading at 77.8%, revealing that current frontier models still fail to comply with observation contracts; (ii) a sharp within-family capability cliff in Qwen 3.5 between 4B (0%) and 9B (56.6%), smoothing to 70.7% at 397B-A17B: what emerges across the cliff is mid-trajectory restraint, not tool-call competence; (iii) non-monotonic scaling across the GPT-5 family: agentic post-training can erode compliance through sycophancy-driven regression; (iv) our failure taxonomy works as an actionable in-context reward signal, yielding +7.1 pp on 42 paired GPT-5.1 failures.
title ContractBench: Can LLM Agents Preserve Observation Contracts?
topic Software Engineering
Artificial Intelligence
I.2.7; I.2.6
url https://arxiv.org/abs/2605.17281