Unified System State and Agent Vitals: Correction and Post-Mortem

This note corrects four records published on 8 April 2026 (10.5281/zenodo.19472715, 10.5281/zenodo.19473627, 10.5281/zenodo.19473857 and 10.5281/zenodo.19473933) and explains how their unsupported claims came to be published. The central claims of all four papers are withdrawn: production validation, new metrics, measured operational impact, mathematical guarantees, and independent confirmation from several methods. Audits and field tests found validation against synthetic data generated around the detectors' own thresholds, known methods presented as new, impact figures that were computed rather than measured, and no true positives for three of five detectors on real agent runs. The work was carried out largely by AI coding and research assistants under the author's direction. Its failures match pitfalls reported for autonomous AI scientist systems. The note closes with lessons for AI-assisted research. A machine-readable table of the August 2026 audit's claim dispositions is attached.