Work

Evidence-Graded Evaluation for Autonomous Agents: A Fail-Closed Methodology

Year: 2026

Type: preprint

Abstract: Working paper v0.1. Proposes an evidence-graded, fail-closed evaluation methodology for autonomous agents: an E0-E4 evidence scale, a gate that refuses to pass on insufficient evidence, scope binding, and negative controls. Includes a case study of a 100/100 self-score versus a 89/100 internal audit traced to an empty test suite and version drift.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.24052276

Author: Zhao, Xinghua

Language: English

Topic: Safety Systems Engineering in Autonomy

Subfield: Safety, Risk, Reliability and Quality

Field: Engineering

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7225578861

Cites: 3

Cited by:

Related to: 10