The Dissipative Generalization Theorem: Overfitting Equals Crypticity Equals Wasted Heat

Staged thematic record of the Viridis Canon (route: S4 (Governance / Alignment)). The Intelligence-Bound spine is unchanged (frozen at v10.0.0, record 20801185); this record links to it via isDerivedFrom the concept DOI 10.5281/zenodo.19317982. Joins the acquisition bound (Goldt–Seifert) and the generalization bound (Xu–Raginsky) on the same quantity I(W;D): decomposing stored information into predictive and cryptic parts gives the identity that the generalization gap equals crypticity exactly — overfitting is crypticity is wasted heat. The generalizing-information rate carries a crypticity debit against the IB ceiling; the optimal learning temperature is β*=1/σ²; efficiency ∈[0,1] equals cos²Θ. Learning-time twin of the Wu-Wei Crypticity Theorem (the Mirror); 17th IB self-application (the Annealer). The 6 core theorems are machine-checked in Lean 4 (Aristotle, zero sorry, axioms ⊆ {propext, Classical.choice, Quot.sound}, statements verbatim and non-vacuous). Scope: the Lean proofs certify the validity of the discrete reasoning, not empirical magnitudes. Working record; paper pending; not peer-reviewed.