# Code of appendices O and P of the benchmark preprint

Preprint: https://eternalsearch.net/omega/benchmark/article · supplementary materials: https://eternalsearch.net/omega/benchmark/supplement

Every script below is the one that produced the published results, with its file paths rewritten to the layout of the archive that carries the same scripts together with their input data: https://eternalsearch.net/omega/benchmark/article-files/2026-11-10-benchmark-appendix-code.zip

- [appendix-text.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/appendix-text.txt) — Builds Appendix O (English) and Appendix П (Russian) of the benchmark supplement from the saved results of the revision after both reviews
- [control-rules.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/control-rules.txt) — Stage 2 of the reanalysis after the first external review (2026-11-02): intervals, within-task analysis, judge panels, self-preference, selection-adjusted estimates, equal-length comparison
- [flash-forecasts.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/flash-forecasts.txt) — Gemini 3.8 Flash in place of Gemini 4 Argon: the three rules frozen in flash-forecast-protocol.json
- [frozen-hypothesis-sample.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/frozen-hypothesis-sample.txt) — Freeze the stratified sample of 100 single-request hypotheses for the blinded manual scoring (decision 1b)
- [frozen-reference-sample.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/frozen-reference-sample.txt) — Frozen sample for the manual check of the reference findings (second review, remark V1)
- [hypothesis-clusters.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/hypothesis-clusters.txt) — A10: recompute the convergence (cluster) fields over the four generators of the paper and report match rate by cluster size and by position
- [manual-audit.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/manual-audit.txt) — Reproduce the completed manual checks from saved, de-identified scores; no model calls
- [manuscript-numbers.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/manuscript-numbers.txt) — Numbers the revised manuscript needs beyond control-rules.py and second-review-analyses.py (revision after both reviews, T6 first-listed hypothesis as a selection rule on all four selection sets, including the Opus repeat generation: PPV, NPV, sensitivity, specificity; intervals for PPV and sensitivity from 10,000 task resamples with the seed of the selection tables (20261026); the unanimity list minus the first-listed hypotheses on the same resamples
- [publication-checks.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/publication-checks.txt) — Three analyses added for the revision of 10 November 2026, computed from the published files only
- [question-cue-recoding.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/question-cue-recoding.txt) — Summary of the blind re-coding of the question cue (second review, V9)
- [reference-reextraction.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/reference-reextraction.txt) — Decision 2: reference findings re-extracted by Gemini 4 Argon High with Sol's original prompt (annotations/reextract) and compared with Sol's findings by Grok 4.7 (annotations/equivalence)
- [rejudge-against-new-references.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/rejudge-against-new-references.txt) — Summary of the re-judging against Gemini's reference findings (second review, V2)
- [review-analyses-selection.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/review-analyses-selection.txt) — Reanalysis of the benchmark preprint after the first external review of 2026-11-02
- [second-review-analyses.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/second-review-analyses.txt) — Analyses for the second external review of the benchmark preprint (3 November 2026)
- [vote-funnel-figure.txt](https://eternalsearch.net/omega/benchmark/article-files/appendix-code/vote-funnel-figure.txt) — Figure 3 of the benchmark preprint: hypotheses retained at each minimum number of "yes" forecasts from the four predictor models, for the Grok, Gemini and Opus selection sets