In plain words: the decay-watchers shipped. Eleven instruments that answer the operator's business question in its live form โ "my feature was independent yesterday; is that verdict rotting right now?" โ went onto the certification lab's shelf. The guaranteed-alarm detector leads (its false-alarm rates are theorems, not tuning choices), followed by the anytime-valid monitors, the 1954 classic entering with an explicit job description (run first as the baseline; every successor must beat it or go home), the fat-tail-native quantile trackers, the offline historian that dates decay retrospectively, and three modern mechanisms with hand-built trust checks the lab must confirm on real data. One entry carries a diplomatic note: the conformal martingale is something the lab has already designed but not built โ the batch says coordinate, don't duplicate. With this delivery, five batches and fifty-six instruments have crossed the interface, every screened queue is empty, and exactly one boundary has never had a dedicated harvest: regime-scoping โ the instruments that say WHICH regimes a conditional feature works in. It gets its first expedition next.
Batch scoreboard โ the interface after five deliveries
Iteration 23 = COND-scoping's first dedicated harvest โ the last cold boundary. Family 16 (tail/extreme dependence) is entirely unswept and holds the parked EVT leads: tail-dependence coefficients, extremograms, extreme-value copulas, the multivariate-extremes change-point line (arXiv:2011.05067), dynamic conditional tail dependence (arXiv:2606.16840), and MS-vine copulas. These are the instruments that name WHICH regimes a conditional feature survives โ the boundary's k_v+G11 incumbents count regimes but don't characterize them. A families-sweep audit rides along (flagged at row 25).
Steering point: no directive = COND-scoping harvest next firing. Directives are ledger-recorded verbatim before taking effect.