# CHANGELOG

## v0.6 — 2026-08-14

### Added
- meta-level binary operator subset selection;
- repeated-subexpression / macro library learning;
- intervention-driven transition model estimation;
- trajectory quotient from learned transitions;
- belief-state task quotient under observation aliasing;
- joint partner/task non-identifiability benchmark;
- causal intervention symmetry-breaking benchmark;
- SciPy HiGHS linear-programming contract repair;
- learned failure-diagnosis decision tree;
- cost-aware protocol-regret objective;
- HIPG-CERT-0.6 semantic invariant preventing behavioral success from being promoted to unique semantic identification;
- theorem-style finite-results note.

### Preserved
- all v0.5 canonical cases;
- exact-bit and Fano lower-bound behavior;
- counterfactual relaxation lineage;
- SUCCESS / INFEASIBLE / UNKNOWN separation;
- SymPy SAT contract bridge;
- binary-reward hypothesis construction.

### Not claimed
- no universal grammar induction;
- no B-TSDPC proof;
- no general latent-ontology identifiability;
- no general transition-learning convergence theorem;
- no general POMDP solver;
- no universal learned impossibility classifier.