v6: results

results/leaderboard.csv: vendor-reported blind accuracy for the GPT-6-Astra and Claude Fable 5.1 agentic runs (18.69 / 20.71)

results/leaderboard.csv: add Claude Opus 5 agentic run (video 69.9, blind 23.6 over 4 permutations, 840 items; same CLI protocol as the vendor returns)

Add certified BenchCheck evaluation logs: COMPLETE.json

Add certified BenchCheck evaluation logs: archives/supporting-00001.tar.gz

Add certified BenchCheck evaluation logs: archives/fable_v2-00001.tar.gz

Add certified BenchCheck evaluation logs: archives/fable_resume1-00001.tar.gz

Add certified BenchCheck evaluation logs: archives/astra-00001.tar.gz

Add certified BenchCheck evaluation logs: README.md

Add certified BenchCheck evaluation logs: OMISSIONS.jsonl

Add certified BenchCheck evaluation logs: MANIFEST.json

Add certified BenchCheck evaluation logs: FILES.jsonl

Add certified BenchCheck evaluation logs: EXCLUSIONS.json

results/leaderboard.csv: final v5 leaderboard (10 models x 840 items, fixed 1 fps / 512-frame protocol) plus the two agentic returns of 2026-09-18 (ddagger rows; no blind run)

Add certified BenchCheck evaluation logs: COMPLETE.json

Add certified BenchCheck evaluation logs: summary.json

Add certified BenchCheck evaluation logs: summary.csv

Add certified BenchCheck evaluation logs: fable_answers.csv

Add certified BenchCheck evaluation logs: category_scores.csv

Add certified BenchCheck evaluation logs: astra_answers.csv

Add certified BenchCheck evaluation logs: README.md

Add certified BenchCheck evaluation logs: MANIFEST.json

Add certified BenchCheck evaluation logs: FILES.jsonl

v5: results