Example of Agents in the Reflexive Laboratory: Trace Analysis of Claude's Work on Zeta

Anthropic published a trace of Claude’s work on a difficult zeta-function research problem, providing an unusually rich example of research-agent transparency. Rather than releasing only a final result, the published materials expose several layers of the research process and make the development of the work available for subsequent analysis. This archive uses that trace as a case study of agents in the Reflexive Laboratory. It preserves the source materials alongside a growing set of derived research objects: transcript mapping, documentary reconstruction, quantitative analysis, sourcebook and storyboard materials, independent reviews, controlled revisions, focused rereviews, precedent and disclosure analysis, software assurance, evidence supplementation, and governance records. The result is a layered trace-analysis archive: Anthropic’s published Claude materials provide the underlying agent trace, while the Reflexive Laboratory adds analytical, review, reconstruction, and governance layers that make the case progressively more inspectable. One important limitation remains: the available source materials do not provide a complete lean-file representation of the agent’s operative state. The archive therefore preserves what can be reconstructed from the published trace while keeping that missing layer explicit.