Iteration 34 — ADMIT #7: IPP Distributional Invariance ADMIT (FOREX) · 27 SHAPE-SHIFTERS

In plain words: the shape-versus-average question got its answer, and it's emphatic. Both tests were run in exact lockstep — same data, same folds, same shuffles — differing only in what they look at: the average relationship (what every certified instrument sees) versus the whole predictive shape (what this candidate sees). The census found 27 features where the average relationship holds steady across all ten regimes while the prediction's shape visibly shifts — and not marginally: on bid_ask_imbalance, the mean-test reads "steady" at evidence level ~3 while the shape-test screams "shifted" at level ~65,000. These are exactly the features a mean-based STABLE certificate would wrongly bless while their risk profile mutates regime to regime. Redundancy check: comfortable — 0.76 at worst against the stack, near-zero against the ancestry dial. Seventh certificate issued: the stack can now see the tail-risk axis of stability. Notably, this candidate needed no rescue margins — the cleanest admission since the regime counter.