Scientific evidence supplement for Evidence-preserving substitution in scientific computation

This scientific evidence supplement supports the manuscript "Evidence-preserving substitution in scientific computation". Evidence-bearing action objects organise native computations with declared scientific uses, current evidence and applicability conditions, making explicit which complete analyses can be supported and what is required to maintain or extend that support. The collection contains frozen protocols, source code, raw and derived evidence, saved-record checkers, independent references, V58 Supplementary Information, scientific figures and editable figure sources. It covers Cell2Fire spatial evidence and dependency-aware grid reuse, PhysioNet DTW retained representations, evidence acquisition and certificate lifecycle, Heat checked inference and false-decision lifecycle, and supporting probability and historical cost audits. Key comparisons distinguish information retention, executable inference and complete delivery. Under the declared DTW interval-retention policy, retained evidence resolves 70% of individual answers but supports only 5% of complete candidate sets at an added threshold; acquiring 108 missing native distances restores all 20 sets. Valid simulation grids support new target questions without another simulation. A stronger checked rule increases Heat certificate support from 190 to 240 judgments using the same enclosure; both complete routes already resolve all 240 judgments. These are separate fixed-panel protocols, not pooled independent samples. Original failures, unsuccessful extensions, verifier amendments and corrected audit records are preserved. Scientific validity and complete-route cost are assessed separately; dynamic, retention and acquisition protocols retain their declared cost boundaries. This supplement complements the five related historical deposits and contains no main manuscript, cover letter or private editorial reports. Reproduction: download all 19 ordered .part-NNNN files together with REASSEMBLE_AND_VERIFY.py and SHA256SUMS_DOWNLOAD.txt into one directory, then run the Python reassembly script. The reconstructed archive contains 151,115,578 bytes and has SHA-256 c166e541cfc8eba60cd7e281cfdd906d44a048682fe73c4b0dcdd90dea88a035. After extraction, begin with reproducibility/REVIEWER_QUICKSTART_V58.md and verify the deposited manifest. Release provenance: this deposit preserves the frozen V58 scientific evidence snapshot. Its original scientific-source files are unchanged in the later V60 manuscript preparation; later revisions update explanatory documents and the conceptual illustration. The Supplementary Information and reviewer instructions enclosed in this archive belong to V58; the current manuscript is accompanied by its separately supplied Supplementary Information. References inside the frozen archive to the earlier reserved record 23845134 and to publication not yet being verified describe the preparation history. Cite the DOI assigned to this record after publication. The frozen archive and its original checksums remain the scientific snapshot; this record's metadata identify its actual publication location. Corresponding authors: Shuyin Xia ([e-mail]) and Guoyin Wang ([e-mail]). Funding: National Natural Science Foundation of China (62221005 and 62222601), and National Key Research and Development Program of China (2025YFF0514700). Licensing: CC BY 4.0 applies to author-created documentation and data where the authors hold the necessary rights. Existing source-code licences and third-party terms remain unchanged; see LICENSES_AND_PROVENANCE.md. In particular, the record-level licence does not replace upstream software licences.