Public reproducibility core for "Exposure is not manifestation: measurement target and output resolution jointly determine which behavioural-faithfulness evaluator wins"

Public, claim-complete reproducibility core for the Scientific Reports manuscript "Exposure is not manifestation: measurement target and output resolution jointly determine which behavioural-faithfulness evaluator wins." The release contains the 720 paired evaluation replies and both estimand labels, per-item scores for the three instrument configurations, canonical aggregate outputs, the matched-geometry result, analysis and figure-generation scripts, and aggregate sufficient statistics that exactly reproduce the reported human Fleiss kappas without disclosing individual ratings. It recomputes the reported tables, figures, paired-cluster intervals, error association, combination-rule metrics, and human agreement statistics. The fixed checkpoint, implementation source, tokenizer, 90,314-record labeled held-out set, training corpora, generation/adjudication prompts, individual human ratings, raw rater files, internal logs, and unrelated project assets are not included. The checkpoint, tokenizer, and pinned source bundle remain available to reviewers on request during peer review and are scheduled for gated-public release at publication. The prior v4 record remains preserved as a restricted provenance archive. Data and documentation are licensed under CC BY 4.0; code is licensed under the MIT License. Patent and trademark rights are not granted. See the included LICENSE.md and PATENT_NOTICE.md.