Wu Wei Dominance: Why Rate-Matched Intervention Strictly Out-Diversifies Aggression in Ecosystem Restoration and Value Formation

Standalone thematic-series record of the Viridis Canon (S4 Stewardship/Governance × S5 Corridors bridge). The core Intelligence-Bound canon spine is unchanged (frozen at v10.0.0, record 20801185); this record links to the spine via isDerivedFrom the concept DOI 10.5281/zenodo.19317982. We prove that, in a community evolving under multiplicative (replicator / Bayesian) drift with an extinction threshold, a sub-critical (slow, “wu wei”) intervention policy attains strictly greater long-run species richness than a super-critical (aggressive) policy. The mechanism is geometric: the extinction face of the abundance simplex is an absorbing boundary — a type driven to zero abundance stays at zero, exactly as Bayesian conditioning cannot resurrect a probability-zero hypothesis (Cromwell’s rule). A single super-critical event drops the support permanently; a trajectory that stays above threshold preserves it intact. Because species richness (the Hill number of order 0) is the diversity ceiling, strict support collapse is strict diversity loss. The discrete dominance result is machine-checked in Lean 4 (Aristotle, zero sorry, axioms ⊆ {propext, Classical.choice, Quot.sound}): the absorbing boundary (Drift.support_subset), permanence of collapse (support_antitone, richness_antitone), super-critical collapse (fast_collapse), sub-critical safety (support_evolve_eq_of_safe), and the capstone wu_wei_dominance_of_rates: richness(evolve Dfast thr T c0) < richness(evolve Dslow thr T c0). The conclusion is a strict inequality between concrete naturals and the hypotheses are satisfiable, so the theorem is non-vacuous. The honest boundary. Which rate regime a policy occupies (the hsub/hsuper hypotheses) is governed by a Kramers / Freidlin–Wentzell escape law, MFPT ≈ exp(ΔU(r_eco − r_int)/σ²), with a critical rate r* = Θ(r_eco) (empirically ≈ 0.85·r_eco — collapse begins below the ecosystem’s own rate). Mathlib lacks Freidlin–Wentzell, so that law is supplied as the hypotheses and deliberately not formalized; everything downstream of it is proven. The management corollary is margin, not matching: because the cliff is exponential, matching the rate is already unsafe; operate with margin below r_eco. Transfer. Under the relabeling extinction → value lock-in, the same theorem says an AI optimizer updating faster than r* ≈ Θ(r_eco) undergoes irreversible lock-in, so corrigibility contains sub-critical updating with a margin — a power-meterable, proxy-independent constraint. Scope. The Lean proofs certify the validity of the discrete reasoning, not empirical magnitudes; the paper develops the physical and AI-safety interpretation. Working paper; not peer-reviewed.