aiqbench analysis
This research addresses a critical failure mode in multi-agent systems where LLMs hallucinate results for commands that were never executed, posing severe risks in high-stakes scientific environments. The authors introduce APEXA, a framework with deterministic execution-integrity enforcement that validates tool outputs against actual compute events rather than relying on textual transcripts alone. By deploying this guard at a major synchrotron facility, the work demonstrates how to safely collapse expert-bound data reduction pipelines using language models without compromising instrument safety or data integrity.
APEXA: Execution-Integrity Enforcement for Multi-Agent LLM Automation of Synchrotron Data Reduction
arXiv:2609.24165v1 Announce Type: new Abstract: Synchrotron data reduction, detector calibration followed by azimuthal integration of terabyte-scale diffraction series, is a multi-step, expert-bound bottleneck that increasingly limits the science rate of user facilities. LLM agents promise to collapse it, but driving a real pipeline with a stochastic model creates a failure mode chat benchmarks cannot see: an agent can report a calibration that was never computed. Correctness here is a property of what executed, not of the transcript. We present APEXA, a deployed multi-agent framework (61 tools over heterogeneous compute, run as a single reasoning loop) automating calibration and integration from natural language at a major light source. We make three contributions. First, execution-integrity enforcement: a deterministic tool-layer guard that refuses to surface any result not backed by an executed tool call, with a parser tolerant of cross-model tool-call format drift: in deployment, a frontier model fabricated a complete calibration-comparison report for commands that never ran, which the guard converts to an explicit non-result; the same code gates an optional motor-control surface at 0/200 adversarial violations against a simulated IOC, versus 15/200 for an equivalent safety prompt. Second, we release APEXA-Bench, an evaluation harness of 58 facility tasks (50 base plus an 8-task cross-detector slice) organized by a four-class physical-consequence taxonomy, the first benchmark axis we know of separating a wasted compute cycle from a damaged instrument; its cross-detector grading against NIST-traceable lattice constants surfaced two latent pipeline bugs. Large-scale agent scoring is left to a full-length study. Third, we validate APEXA on real beamline data: from one natural-language prompt it recovers detector geometry and integrates a full attenuation/exposure sweep. We release the framework, harness and traces.
Abstract:Synchrotron data reduction, detector calibration followed by azimuthal integration of terabyte-scale diffraction series, is a multi-step, expert-bound bottleneck that increasingly limits the science rate of user facilities. LLM agents promise to collapse it, but driving a real pipeline with a stochastic model creates a failure mode chat benchmarks cannot see: an agent can report a calibration that was never computed. Correctness here is a property of what executed, not of the transcript. We present APEXA, a deployed multi-agent framework (61 tools over heterogeneous compute, run as a single reasoning loop) automating calibration and integration from natural language at a major light source. We make three contributions. First, execution-integrity enforcement: a deterministic tool-layer guard that refuses to surface any result not backed by an executed tool call, with a parser tolerant of cross-model tool-call format drift: in deployment, a frontier model fabricated a complete calibration-comparison report for commands that never ran, which the guard converts to an explicit non-result; the same code gates an optional motor-control surface at 0/200 adversarial violations against a simulated IOC, versus 15/200 for an equivalent safety prompt. Second, we release APEXA-Bench, an evaluation harness of 58 facility tasks (50 base plus an 8-task cross-detector slice) organized by a four-class physical-consequence taxonomy, the first benchmark axis we know of separating a wasted compute cycle from a damaged instrument; its cross-detector grading against NIST-traceable lattice constants surfaced two latent pipeline bugs. Large-scale agent scoring is left to a full-length study. Third, we validate APEXA on real beamline data: from one natural-language prompt it recovers detector geometry and integrates a full attenuation/exposure sweep. We release the framework, harness and traces.
5 pages, 4 figures. Accepted at the 4th TPC Workshop @ SC'26 (Building Open AI Infrastructure, Models, and Agentic Systems for Science). Code and benchmark: this https URL ; data: this https URL
Artificial Intelligence (cs.AI)
I.2.11; I.2.7; J.2; D.2.4
arXiv:2609.24165 [cs.AI]
(or arXiv:2609.24165v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.24165 arXiv-issued DOI via DataCite (pending registration)
Full text under the publisher's open licence · 326 words
Related coverage
[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space · Oct 13, 2026
AgentOps: Operationalize agentic AI at scale with Amazon Bedrock AgentCore
AWS · Jun 1, 2026
Why do models task game?
AI Alignment Forum · Aug 6, 2026
[AINews] not much happened today
Latent Space · Sep 18, 2026