# CHANGELOG

## v0.9 — 2026-08-14

### Added
- finite intervention-language learning over 32 behavioral kernels;
- multi-variable, non-invertible, stochastic, and composite intervention operators;
- finite grounding-number computation ($G^*=2$ on an $S_3$ gauge class);
- active POMDP discovery of $P,O,R$ without hidden-state resets;
- operator-family growth across unary/binary/ternary, multi-valued, and sequential families;
- HIPG-TEMP-PROOF-0.9 replayable temporal proof objects;
- theorem-anchor registry with assumptions, scope, checker ID, version, and hash;
- online drift detection before operator retirement;
- HIPG-EQUIV-0.9 public equivalence-class artifacts.

### Preserved
- all 50 v0.8 canonical cases;
- SUCCESS / INFEASIBLE / UNKNOWN separation;
- permanent semantic equivalence classes are not collapsed to a convenient representative;
- lower-bound and counterfactual branches remain distinct.

### Not claimed
- no B-TSDPC proof;
- no universal intervention language;
- no general POMDP identifiability theorem;
- no general minimal grounding theorem;
- no unrestricted temporal theorem prover.