Iteration 18 — Provenance debt cleared 4 CITATIONS RESOLVED · 3 ROWS · 79 TOTAL B-05 WITHIN REACH

In plain words: a housekeeping firing that pays compound interest. First, the campaign's oldest debt is cleared: four instruments had sat unscreenable since the very first day because their papers were never formally resolved — the no-hearsay rule held them at the door for fourteen iterations. All four citations resolved today (adaptive conformal prediction, the original Bayesian changepoint paper from 2007, the discriminative drift detector, the drift-adaptive ensemble), unlocking the EXPIRY boundary's path to its first handoff batch. Second, the invisible infrastructure got its citations: nearly every instrument in this catalog says "needs a dependence-aware correction" — the two standard ways of doing that (the 1987 analytic variance fix and the 2010 multiplier bootstrap) are now catalogued with sources and implementations, alongside the standard offline changepoint segmenter (an actively maintained library, distinguished on the record from the parametric test the lab boneyarded). One small honesty note: the multiplier bootstrap's citation was first attempted with a wrong identifier from memory — the failure and the recovery are both on the ledger. The web-search channel answered a library-phrased query and contributed five leads.

 Scaffold day (July 4):  H-011 ACI · H-012 BOCPD · H-013 D3 · H-014 PWPAE
                         entered with checked CODE but unresolved PAPERS
                         → "provenance completion pending" → unscreenable
 14 iterations later:    all four primary sources resolved (arXiv ×3, Crossref ×1)
                         → screenable at last → EXPIRY cohort now 5 + 5 prior passers
                         → screen S-06 could reach 10 → B-05 unlocked