Operational Transition Test (OTT): Experimental Crucible v0.2 — Prospective Results (H1 = FAIL)

Abstract

This record contains the prospective results of the Operational Transition Test (OTT) Experimental Crucible v0.2.0. The protocol, hypotheses, thresholds, benchmark families, development/evaluation split, resource accounting rules, and failure criteria were publicly frozen before execution under DOI 10.5281/zenodo.22116064 (public freeze timestamp 2026-08-26T17:43:22Z; run authorization 2026-08-26T17:56:20Z). The full OTT achieved high overall classification performance on the held-out evaluation set (balanced accuracy 0.9723214286; positive-family TPR 1.0000; overall FPR 0.0553571429), but the primary frozen hypothesis H1 FAILED because the family-specific false-positive rate for the context_fragile_macro negative family was 0.3750, exceeding the predeclared per-family limit of 0.10. H1 = FAIL No thresholds were changed after observing development or evaluation results. No post-hoc retuning was used to convert the failed result into a pass. The failure localizes a specific limitation of the current formulation: held-out operational gain does not by itself guarantee reliable cross-context transfer of the declared macro-function. Secondary (non-decisive) findings. The full OTT outperformed all frozen simpler variants in balanced accuracy; the best reduced variants (F+K+E and F+K+E+D+R) reached 0.9276785714, leaving a full-vs-best-simple margin of 0.04464, below the frozen auxiliary superiority margin of 0.05. F+K+E and F+K+E+D+R produced identical classification performance on this benchmark: v0.2 did not demonstrate independent discriminative value for D and R beyond F+K+E in this benchmark. This record is published as the unmodified prospective outcome of the frozen v0.2 protocol and is intended to inform a subsequent, separately frozen revision (v0.3) rather than retroactively modify the current test. This is a synthetic-benchmark result; no cross-domain or ontological claim follows from it. Contents. The ZIP is the canonical complete result artifact (run authorization, development freeze, evaluation lock, predictions, ground truth, scoring, result summary/report, integrity manifests). RUN_RESULT_REPORT.md is included unchanged for human-readable preview. Lineage. Results of the publicly frozen prospective protocol DOI 10.5281/zenodo.22116064; theoretical predecessor: Gątkowski (2026), When Output Becomes Infrastructure, v0.1.0, DOI 10.5281/zenodo.22115305. Licensing. Record license CC BY 4.0. Software components inside the package remain under Apache License 2.0 according to the research-line licensing policy (LICENSES.md of the frozen protocol record). Copyright © 2026 Sławomir Grzegorz Gątkowski. Integrity (SHA-256). Results ZIP: b2f2b14acc0a61384672d63511c0f48d8fe6adf6aff1297b0efcbe8276ab2796 — verified before publication; uploaded files confirmed byte-identical.