Current-evidence operation-choice supplement for Evidence-preserving substitution in scientific computation

This supplement archives the uploaded V83 bundle for the current-evidence operation-choice experiment associated with "Evidence-preserving substitution in scientific computation". The scientific experiment is the frozen V82 snapshot; subsequent manuscript revisions refine exposition and display organization without changing the scientific executions. Archived V83 presentation and document-mapping files retain their original scope; current submission figures and mappings may be supplied separately with the manuscript. A rule developed on eight INCART development sources was frozen before signal access to twelve prospectively fixed records from the MIT-BIH Normal Sinus Rhythm Database (NSRDB). Current valid exact evidence is reused; unresolved comparisons use guarded DTW bounds, a fixed state-dependent choice rule and native exact-distance acquisition to establish complete current candidate-ID sets. The scientific package records the frozen design and code, source receipts and selected waveform fragments, development results, routed outputs, complete-route timing and a separately implemented post-route reference audit. Local pre-signal freezing is not external preregistration, and independent implementation is not external-team replication. All four routes complete 960 analyses across four threshold menus starting with empty evidence caches and 2880 analyses across a twelve-stage lifecycle per repeat. Relative to the matched fixed cascade, the current-evidence chooser reduces measured complete-route time by approximately 3.8% and 4.2%; simpler forward-screening and exact-cache routes remain faster. Fifteen timing repetitions are technical repeats; the source-level comparison uses twelve records. These results concern computational waveform retrieval, not clinical discovery, global scheduling optimality or a general efficiency guarantee. Reproduction: extract the bundle and the contained scientific archive, then run python reproduce_operations.py verify from dtw_operation_choice_20261103/. The scientific snapshot used by the current manuscript is V82_Operation_Choice_Reproducibility.zip (109069857 bytes; SHA256 e51259d9ab10b6b7dbc8e7eeabe3e80f1de6a6074e592107a632de43cb9b2388). This hash identifies that inner scientific archive, not the uploaded outer bundle. Public-file and nested-archive verification is performed separately from the scientific audit. Data attribution and rights: NSRDB v1.0.0 (George Moody, 1999), DOI 10.13026/C2NK5R, and the INCART development source (Evgeny Yakushenko, 2008), DOI 10.13026/C2V88N, are distributed by PhysioNet under Open Data Commons Attribution License v1.0. Retain all original notices and citations. Data licensing does not assign a license to independently written code, text or figures. Component-specific terms supplied in the archive remain applicable; this record does not grant additional rights for components without an explicit license. Funding: National Natural Science Foundation of China, grants 62221005 and 62222601; National Key Research and Development Program of China (2025YFF0514700). Joint corresponding authors: Shuyin Xia and Guoyin Wang.