โ€บNavigation

Iteration 30 โ€” ADMIT #5: seqICP, the Env-Free Corroborator ADMIT (FOREX)

In plain words: the deciding match. Both instruments โ€” the reigning identifier and the label-free challenger โ€” were run over the same 130 forex features on the same ten regimes, and their readings compared feature by feature. On forex the challenger's readout came alive in a way it never did on crypto: a genuinely graded scale rather than all-or-nothing. The redundancy verdict: correlation 0.65 with the reigning identifier โ€” related, as two honest invariance instruments should be, but nowhere near the 0.95 "same gauge twice" kill bar. And the disagreements are exactly where the value lives: fourteen features only the identifier flags, three only the challenger flags, and on the one feature both identify, the answers overlap sensibly. Fifth certificate issued โ€” scoped to forex, in a role nothing else in the stack holds: a second, mechanically independent opinion at the STABLE boundary that needs no regime labels at all โ€” insurance for exactly the case where our regime definitions themselves are wrong. Its crypto persistence role stays excluded (yesterday's kill stands untouched). The queue advances: next candidate, H-023.

Preflight (A0): 2026-07-10 โ€” load1 5.12/32c (โ‰ค24) ยท 23 GiB avail ยท CH + sidecar active ยท no heavy jobs โ†’ ALL PASS. One capped run (2.97 min), committed forex grid + sha-pinned cells, permutations of real rows only.

The head-to-head (130 subjects, same grid, same conventions)

ElementReading
seqICP readout profile (forex)GRADED: n_acc 0 โ†’ 100 feats ยท mid-scale โ†’ 7 ยท full-accept โ†’ 23 (vs the bimodal crypto profile โ€” the substrate matters)
M3 attackworst |ฯ| = 0.653 (s_max_p vs e-ICP n_acc; positive, as related invariance instruments should be) โ€” far under the 0.95 kill bar. Crypto M3: 0.252 (row 37). Non-redundant on both substrates.
Disagreement structure (the role evidence)e-ICP-only identifications: 14 features ยท seqICP-only: 3 ยท both: 1 (intra_intensity_qps_q1000 โ€” seqICP {duration_us, intra_intensity_qps} โŠƒ e-ICP {duration_us}; overlapping, e-ICP more specific)
Ladder summaryM0/M1 โœ“ (screening + JASA lineage + exam) ยท M2 โœ“ (block-null standing rule) ยท M3 โœ“ (both substrates) ยท M5โ€ฒ โœ“ (entrance exam, row 36) โ€” the row-34 M3+M5โ€ฒ precedent ยท crypto M4 ran and FAILED โ†’ crypto scope stays EXCLUDED (row 38)
Verdict (row 39)ADMIT(forex, 10-env EURUSD@5 grid, role = env-free corroborating invariance readout โ€” mechanically independent second opinion at the STABLE boundary, works where env labels are wrong or absent, expiry = next completed forex regime slice) โ€” ADMIT #5. Feed row H-016 consumed.

What the stack gains: STABLE claims on forex can now cite two mechanically unrelated invariance instruments (env-labelled e-value fold-products + label-free time-block statistics). Agreement between them is strong evidence; divergence is a flag for regime-definition problems โ€” a failure mode nothing else in the stack could see.

โ–ถ Next iteration

Iteration 30 ยท 2026-07-10 ยท ADMIT #5 (row 39) ยท capped ยท zero synthetic data ยท append-only ยท evidence: seqicp_forex_head_to_head.py ยท seqicp_forex_head_to_head_results.json