Iteration 7 — Family 2 swept 9 HARVESTED · 59 TOTAL FAIL SCREEN COHORT READY

In plain words: this firing harvested the family of instruments that answer a question every trading system eventually asks: "how do I keep checking a claim while the data keeps arriving — without fooling myself by peeking?" Classical statistics breaks when you look at accumulating evidence repeatedly; this family (the "anytime-valid" school) is built so the guarantee holds no matter how often you look or when you stop. Nine instruments entered: change-detectors with provable false-alarm control, monitoring bands that stay honest forever, a technique for testing hypotheses when the usual textbook conditions fail (exactly the mixture-distribution mess that killed a classical test on our data), sequential versions of independence and regime-homogeneity tests, and quantile trackers that are naturally immune to fat tails. Two entered without dedicated code but with constructions cheap enough to state on the record. One (the conformal test martingale) was flagged as already sitting in the admission lab's own to-build queue — the flag travels with the entry so nobody duplicates work. Two theory papers were deliberately left OUT — they would have padded the count without adding a usable gauge. The FAIL boundary now has eleven screenable instruments: its first interrogation batch is due next firing.