# CHANGELOG

## v0.7 — 2026-08-14

Added derived-operator proposal, Dirichlet active dynamics, learned observation models, first-class latent equivalence classes, active class splitting, OOD abstention, mixed SAT+LP bridges, clause-level provenance anchors, meta-regret, and HIPG-CERT-0.7 semantic checks.

Preserved all 34 v0.6 canonical cases and the SUCCESS / INFEASIBLE / UNKNOWN separation.

Not claimed: B-TSDPC proof, universal grammar induction, general causal discovery, unlabeled hidden-state observation learning, or general SMT/Lean verification.