Methodology Note 01 v4. 0 — Amendment to the Pre-Registration (DOI 10. 5281/zenodo. 20125967) of the Marin T. Kael Citation-Fidelity Research Programme. v4. 0 is an infrastructure migration, not a methodological reformulation. The score-algorithm, question-set, and aggregation logic are byte-identical to v3. 0 (DOI 10. 5281/zenodo. 20262362): 27 of 27 historical datapoints match 1: 1 between the v3. 0 worker-JS implementation and the v4. 0 Python implementation. What changed: the measurement instrument moved from in-Worker JavaScript score-functions to a standalone Python package on top of EleutherAI lm-evaluation-harness v0. 4. 12. The harness is now reproducible on any researcher's laptop in under five minutes with pinned tooling versions and a public GitHub repository. Change-Discipline (locked): patch (v4. 0. x) = bug-fix without methodological impact, no Amendment DOI; minor (v4. 1) = additive change (new NEGATIVEHALLU pattern, new model adapter), Amendment DOI required, task stays marinᵣesearchᵥ1; major (v5. 0) = breaking change (new question, score-formula shift, anchor redefinition), new task marinᵣesearchᵥ2, full Pre-Registration Amendment, all post-v4 data treated as separate cohort. The accompanying implementation (lm-evaluation-harness tooling, custom model adapters, CI cron, snapshot-to-D1 push) is published as software v0. 4. 2 (DOI 10. 5281/zenodo. 20364157).
Marin T. Kael (Sun,) studied this question.