In plain words: interrogation round for the EXPIRY boundary โ the instruments that watch a live verdict rot and raise the alarm. Of the ten catalogued, only six could legally face the gates: four scaffold-era entries still lack a verified primary paper (the campaign's no-hearsay rule bites its own early work โ they're named, not screened, and resolving their citations is now a listed task). Five of the six passed, four of them with guarantees printed in their own papers โ false-alarm-rate control, forever-valid bands, a betting martingale's built-in bound, and fat-tail-proof quantile tracking. The framework paper that spawned this whole family went out as expected: frameworks inspire instruments but aren't instruments (the same principled exit as the theory anchors before it). The headline is the number that DIDN'T happen: five passers is below the ten needed for a handoff batch, and the rulebook forbids padding โ so the shortfall is reported as a finding, with the exact refill path named. Separately, the web-search channel (trading-phrased, so it answered) surfaced one new lead: a 2026 regime-performance-persistence framework already validated on crypto factors.
The screen
10 EXPIRY entries
โโ 4 UNSCREENABLE: H-011 ACI ยท H-012 BOCPD ยท H-013 D3 ยท H-014 PWPAE
โ (scaffold-era, primary papers still unresolved โ the no-hearsay rule
โ applies to our own early rows too; provenance-completion task listed)
โผ 6 face M0/M1
1 OUT (OFF-GOAL): H-003 SAVI โ the FRAMEWORK; its four concrete instruments
are catalogued separately (theory-root retention)
โผ
5 SCREENED-M0M1-PASS ยท below the 10-entry feed threshold
โ NO B-04 โ shortfall reported per the quantity-honesty rule
โ refill path: resolve the 4 pending provenances + families 3/5/6/18
The five passers
id
Instrument
M1 anchor
H-051
E-detectors (stcpR6)
nonasymptotic false-alarm bound โ strongest in cohort
H-052
Time-uniform confidence sequences (confseq)
time-uniform coverage (paper-native)
H-054
Conformal test martingales
Ville's-inequality bound; Frontier-#4 flag travels
Iteration 14 = FRAGILEโCONDITIONAL harvest (deficit 54 โ the largest actionable gap). STABLE's deficit (67) is bigger on paper, but its serving families are all swept โ fresh STABLE yield needs lead-verification work, while FRAGILEโCOND has three entirely unswept families: 9 (feature-selection stability indices โ Nogueira, Kuncheva, top-k overlap), 13 remainder (MMD two-sample, KS/CvM block-calibrated, local Gaussian correlation lead), and 19 (verdict/ranking stability). The MRP lead (AlexanderโFabozzi 2026) also serves this boundary.
EXPIRY refill (alternative slice): resolve the four pending provenances (ACI, BOCPD, D3, PWPAE) + family-5/6 harvest โ puts B-04 within reach in two firings.
Steering point: no directive = FRAGILEโCOND harvest next firing. Directives are ledger-recorded verbatim before taking effect.
Machine-readable record: ledger row 18 + catalog ยงS-04 + taxonomy iter-13 lead (MRP). Pages are append-only.