Heritage response scoring: an image-free fixed-output reproduction core

Abstract

This software and data package supports deterministic recomputation of response parsing and panel admission in a finite multimodal heritage panel, using frozen experimental responses. The package contains 9,600 follow-on raw-response records under aliases M1-M4; a 43,200-row derived evaluation table; the evaluator contract and implementation; executable analysis code; and 15 reference outputs. The reproduction command checks 38,400 follow-on evaluator instances against the derived table and regenerates all 15 reference outputs. The remaining 4,800 pilot-derived rows receive structural checks because their raw-response records are not included. M1-M4 identify frozen response groups, without disclosing exact model lineage or guaranteeing anonymity against inference from the outputs. Version 0.7.0 fixes analysis entry-point loading under Python safe-path settings and adds regression checks for running the tests and reproduction consecutively in the same extracted directory. The tests remove only their own generated bytecode file; the integrity checker continues to reject other undeclared files or caches. The calculation implementation, raw and derived data, protocols, analysis specifications, and all 15 reference outputs are unchanged from version 0.7.0-rc4. Execution has been checked locally with Python 3.12.14 using the standard library, including PYTHONSAFEPATH=1 and python -P. These checks establish deterministic recomputation, not independent scientific validation. The public core excludes source images and masks, weights and controlled identity mappings, individual expert identities or review returns, correspondence, and manuscript files. It does not provide fresh inference, establish expert validity or heritage accuracy, or reproduce every historical and supplementary analysis of the associated study. See README.md for execution instructions, DATA_DICTIONARY.md for the input structure, and RIGHTS_AND_EXCLUSIONS.md for scope and rights boundaries. The archive includes its file-integrity manifest and checksums. RELEASE_STATUS.json records the pre-upload build-time snapshot; the repository record provides the version-specific DOI and publication status after publication. Licensing is allocated by file type. All .py software files, including embedded comments and docstrings, are licensed under the MIT License. All non-.py documentation, data, selection and arrangement, and tables are licensed under Creative Commons Attribution 4.0 International (CC BY 4.0), limited to author-controlled copyright and similar rights and only to the extent the licensor is authorized. These are separate scopes, not a choice of either licence for every file. No ownership is asserted over uncopyrightable content. Third-party rights are not relicensed, and no rights to excluded images, weights or expert materials are granted. The notices do not certify downstream training permission or compliance with upstream third-party obligations. The complete MIT notice and further exclusions appear in RIGHTS_AND_EXCLUSIONS.md.