An audit-grade collaborative review of variational free energy as it is used in active inference, organized as a transparent re-presentation of the central variational identity together with a sentence-level provenance trail and a deterministic reproducibility test suite (87 pytest assertions across 11 numerical demonstrations, CI-verified across Linux × Windows × macOS × Python 3.11/3.12/3.13). Treats Maren's Themesis Technical Report TR-2019-01v6 as a test case for the audit method, not a target. Layer 1 (AI-executable) of the audit remediation plan is complete; Layer 2 (human expert review) gates remain pending. Source materials referenced (Maren TR-2019-01v6; Parr/Pezzulo/Friston 2022; SOURCE C GPT review) are NOT redistributed in this repository.

AI authorship and contribution disclosure: produced by Anthropic Claude (Opus 4.7) in collaboration with Michael Polzin (organic operator) over four documented sessions (audit, drafting, audit-of-audit, P0-P4 remediation). The prior independent peer review captured as SOURCE C was produced by the 'Ai Onna' Custom GPT, running on OpenAI's GPT platform, built on the ORCHESTRATE Method (https://www.amazon.com/ORCHESTRATE-Prompting-Professional-AI-Outputs/dp/B0G2BJKDM6); the 'Jules' Custom GPT (also OpenAI / ORCHESTRATE) is acknowledged as additional AI co-contributor. See Manuscript_Draft_v2.md front matter and Appendix E.2 for full session-by-session provenance. AI-generated content remains provisional until reviewed by qualified human experts.