Research data and minimal replay code for Representation-Dependent Scores in IFC Editing Evaluation: A Controlled Audit and Discrimination Checks

Version 1.0.0 contains research data and minimal replay code accompanying Representation-Dependent Scores in IFC Editing Evaluation: A Controlled Audit and Discrimination Checks. It documents a controlled, version-specific audit of IFC editing evaluation; it is not a preprint or a claim of journal publication. Included: all 15 R1 UPDATE configurations and qualification records for 11 related input files; three R2 default-axis source controls; all eight file-calibration conditions and their 48 metric values; the complete task11 witness; three result tables; provenance and fixed dependency identities. Related inputs and shared helper implementations are not independent engineering projects. Seven baseline UNKNOWN outcomes are retained. Executable coverage is limited to a minimal task11 replay entry, using separately acquired third-party dependencies. R1, R2 and calibration results are archived observations; the full cohort and calibration runners are not included. The prior relocated task11 replay used the same installed engine and is not external replication. Excluded: third-party IFC files (original and derived), upstream scoring source, third-party binaries, natural-generation/consumer pilot originals and API logs, and the manuscript/figure submission files. Fixed URLs and SHA256 identities are supplied where needed. No new model calls, IFC edits or scientific scoring were performed for archive preparation. Licensing: study-derived data, tables and study-written documentation are CC BY 4.0. Study-written Python code is MIT, copyright 2026 YuanQiu Zheng. This is a file-specific allocation, not a choice of either license for every file. See RIGHTS.md and rights_inventory.csv. Third-party dependencies retain their own applicable terms and are not redistributed. The audited BIBIMBAP work is referenced by DOI 10.35490/EC3.2026.271. Fixed source snapshots are specified in the archive; they are not asserted to be the historical execution versions of the published leaderboard. These controlled observations do not estimate natural misgrading prevalence or revise a leaderboard. OpenAI Codex assisted implementation, data extraction, documentation and internal checks; it is not an author or an independent human reviewer.