AE v2 — Extrapolable Structural Signature: Pre-Registered Experimental Specification (Kappa Program / Katashi Principle)

AE v2 — Extrapolable Structural Signature is a pre-registered experimental specification within the Kappa Program. It was written and frozen in its entirety before any result of the experiment it specifies had been observed.The experiment addresses the strongest methodological criticism available against the Katashi Principle (Obsessive Coherence — Volume I, DOI: 10.5281/zenodo.20516886): that the crystallization signature might be an artifact of the measurement apparatus — CALM self-calibration, quantile thresholds, the structural pressure integrator — rather than a property of the observed system ("the principle is the pipeline"). AE v2 tests this directly: a classifier is trained exclusively on calibration-free geometric features (spectral entropy, dominant-eigenvalue share, Forman-Ricci curvature, persistent homology H1 counts, mean correlation, raw structural volatility) of mature financial universes, and evaluated on systems it has never seen, with no local calibration apparatus. If the signature survives the removal of the instrument, the phenomenon is more fundamental than the protocol that measures it.Three pre-declared hypotheses with explicit nulls: H1 — portability of regime recognition across mature universes (leave-one-universe-out; success criterion: median AUC ≥ 0.75); H2 — external market-event prediction on a juvenile cohort (universes whose self-calibration is forced by data-boundary or degenerate-threshold conditions), against frozen external targets and a frozen nested baseline (hazard v2 GBM, test AUC 0.741 at 63 days); H3 — sample-efficiency curve over degraded input windows. Cohorts are nominally frozen in Appendix A (8 mature, 13 juvenile, extracted from the re-anchored production run of 2026-06-09). Architecture, hyperparameters, seeds, temporal embargoes, anti-leakage controls, statistical tests, and the interpretation of every possible outcome — including failure — are pre-committed in the document. Results will be published whatever the outcome, as version 1.1 of this record.The canonical file is AE_v2_preregistration_spec_EN.md.SHA-256: 1201637a7c57bf72e6c06e830faafdaece18c0d70f261a228deb9e934a423cedPart of the Kappa Method ecosystem: Obsessive Coherence — Vol. I (10.5281/zenodo.20516886), Domain-Agnostic Structural Signature paper (10.5281/zenodo.19393239), Kappa-LLM (10.5281/zenodo.18883790), Kappa Method v2 (10.5281/zenodo.19339548).David Ohio | Independent Researcher | odavidohio@gmail.com | CC BY 4.0 ═══ v1.1 — RESULTS (2026-06-11) ═══ This version adds the complete results package of the pre-registered experiment, evaluated strictly against the deposited criteria. H1 — PASSED. Median leave-one-universe-out AUC 0.8974 (criterion ≥ 0.75); above chance in 7/8 held-out universes (criterion ≥ 80%). A student trained exclusively on calibration-free geometry recognizes the structural regime of mature universes it has never seen: the signature survives the removal of the calibration apparatus. H2 — FAILED per the deposited criterion. The transferred student matched or beat the juvenile universes' own self-calibration in only 8/16 comparisons, and added no value over the frozen nested baseline (0.7439 vs 0.741). The pre-committed interpretation (Section 10, row 2 of the specification) applies verbatim: the signature is real but non-portable beyond K0; the juvenile cohort requires graduation, not transfer. Notable finding: juvenile self-calibration is bimodal — for iran_war it reached AUC 0.934 on external 63-day events, while for energy-related universes it was anti-predictive (0.11–0.21). H3 — flat curve. Median LOUO AUC 0.897/0.893/0.897/0.894 for windows of 22/15/10/5 days: within the mature class, recognition is nearly pointwise. Disclosure: pre-registration deposit and experiment execution occurred on the same day (2026-06-10); the timestamp therefore proves order of publication, not order of knowledge. This limitation is owned explicitly in RESULTS_v1_1.md and addressed by a binding prospective commitment: the three frozen students (SHA-256 in provenance_v1_1.json) will be re-evaluated, unchanged, on data strictly posterior to 2026-06-10, on 2026-09-08 (v1.2) and 2026-12-07 (v1.3). Data that did not exist at deposit time is the witness that cannot be forged. Package: 13 files — results report, frozen student weights (3 seeds), reference AUCs, raw results, frozen labels, out-of-fold scores, execution script, figures, and per-file SHA-256 manifest.