โ€บNavigation

Iteration 32 โ€” ADMIT #6: IAS Ancestry Readout ADMIT (FOREX) ยท SCOPED OUTPUT

In plain words: the "two dials on one sensor" fight, settled by taking the metaphor literally: one set of underlying measurements, both combination rules applied to the exact same numbers, across 130 features. The result split the candidate in two. One of its dials (how many smallest steady sets exist) really is a near-copy of the reigning identifier's dial โ€” correlation 0.937, a hair under the 0.95 kill line, and the loop wrote that number down in red rather than hiding it: that dial is not certified. But the candidate's namesake dial โ€” the ancestry reading itself โ€” measures something the identifier doesn't: on four real features it produces evidence where the identifier returns pure silence, and its overall correlation stays a comfortable 0.82. Sixth certificate issued, deliberately narrow: only the ancestry reading is certified, only on forex, expiring at the next regime turn like everything else. A surprise came out free of charge: for 43 of 130 features the feature itself reads steady across all ten regimes before any anchors get involved โ€” a brand-new lens on which features deserve the STABLE conversation at all.

Preflight (A0): 2026-07-11 โ€” load1 3.23/32c (โ‰ค24) ยท 33 GiB avail ยท CH + sidecar active ยท no heavy jobs โ†’ ALL PASS. One capped run (2.9 min), committed forex grid + sha-pinned cells, permutations of real rows only. Decision rule pre-registered in-script before running.

The shared-oracle M3 (one e-value set, two grammars, 130 subjects)

ReadoutWorst correlation vs admitted outputsRuling
n_minimal (count of minimal invariant sets)0.937 vs e-ICP n_acc โ€” 0.013 under the kill barnear-redundant repackaging โ€” NOT certified as an independent output; flagged for expiry review (row-35 precedent)
anc_size (the ancestry reading โ€” the candidate's namesake)0.824 max (vs ident_size); 0.19โ€“0.22 vs everything elsegenuinely non-redundant โ€” certified
Informative gain4 features where ancestry speaks and identification is silent: churn_run_count_q1000, ofi_flicker_adjusted_bar, ofi_tick_count_bar, spread_variance โ€” the paper's raison d'รชtre realized on real data
Verdict (row 41)ADMIT(forex, 10-env EURUSD@5 grid, role = higher-power ancestry readout at the STABLE boundary โ€” invariance evidence where identification returns empty; admitted output = ancestry set/size ONLY, expiry = next completed forex regime slice) โ€” ADMIT #6. Feed row H-023 consumed.

Free bonus (report-only): the lattice's empty set was accepted for 43/130 features โ€” those targets read regime-invariant BY THEMSELVES (no anchors involved). A new, orthogonal lens on the k_v=8/8 population for the coming STABLE-declaration proposal.

โ–ถ Next iteration

Iteration 32 ยท 2026-07-11 ยท ADMIT #6 (row 41) โ€” scoped output, margins in red ink ยท capped ยท zero synthetic data ยท append-only ยท evidence: ias_ladder_m3.py ยท ias_ladder_m3_results.json