Iteration 23 — COND-scoping's first harvest 7 ROWS · 91 TOTAL 16 OF 20 FAMILIES SWEPT

In plain words: the last cold boundary got its expedition. Regime-scoping instruments answer the question the incumbents can't: not "in how many regimes does this feature work?" but "in WHICH regimes — and what do those regimes look like in the tails?" Seven entries: the asymptotic tail-dependence coefficient (what LGC's local fits approximate, taken to the limit), the extremogram — possibly the most substrate-native instrument in the whole catalog, since serial dependence of extremes is literally its object of study — extreme-value copula inference, the dynamic Hüsler–Reiss graphical models whose crash-graph/rally-graph asymmetry was this morning's web-search find (crypto crashes together, rallies apart), and the 2002 exceedance-correlation classic. Two conceptually excellent candidates entered and immediately exited with NO-IMPL stamps: the extremes change-point tracker and the Markov-switching vine copulas — the purest scoping idea in the family, whose building blocks exist in two maintained libraries but whose switching layer nobody has published code for; both carry revisit notes. Also on the record: two wrong-from-memory citation identifiers and their recoveries, and the campaign's first same-firing conversion of web-search finds into verified catalog rows. The families-sweep audit rode along: sixteen of twenty families swept; the four remaining are named with their leads.

 SWEPT (16/20): 1* 2 4 5 6 7 9 11 12 13 15 16 17 18** 19 20
   *family 1: GROW-variant residue noted   **family 18: swept by scope call
 NOT SWEPT (4): 3  conformal family     — EnbPI, SAOCP, conformal-PID leads
                8  conditional independence — CondIndTests, cdcsis leads
                10 resampling machinery — circular block, subsampling leads
                14 predictive-ability   — DM, GW, SPA, MCS (serves CONDITIONAL — untouched)