Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces a "harness" that surrounds agentic scientific workflows, treating output release as a decision governed by mandatory checks, failure containment, version tracking, and a risk model. The harness enforces trust by preventing agents from overriding its release decisions, demonstrating that containment and validation significantly increase correct results, while risk estimation offers efficiency gains over random validation. Crucially, the language model agent provided no measurable benefit, highlighting that structural controls, not agent intelligence, are paramount for trustworthy scientific output.
Agentic workflows can be made trustworthy by enforcing explicit release controls that prevent agents from overriding structural checks, even when the agent itself adds no measurable benefit.
An agentic scientific workflow can run to completion and still release wrong results: a silent numerical error, a stale input, or a defect in a shared upstream step can contaminate several outputs while a sibling branch stays correct. We treat trust in such results as a release decision. A harness around the workflow's directed acyclic graph (DAG) decides which outputs may be used: it enforces mandatory checks, contains failures to the affected branch, recomputes from tracked versions, adds validators, and releases a result only if a risk model estimates its probability of being invalid as low. Language-model agents work inside the harness, choosing corrective actions and reviewing releases, but cannot override it. On 60 held-out paired instances of two workflows with injected faults, we switched components on and off. Each supplied a different property. Containment and recovery raised the number of correct results. Validation supplied safety: fixed rules without validation released 37 invalid results, while validating every candidate released none. Risk estimation supplied efficiency, matching that safety with 31% fewer validators at somewhat lower yield, although better targeting than random validation was not shown. Lineage evidence helped on one workflow. The language model added no measurable benefit: whether invalid results escaped depended on the harness, not the model. This proof of concept points to a practical path: placing agents inside explicit release-control structure can keep released scientific results trustworthy, and it gives more capable agents a measurable baseline to improve on.