Frontier subset (swe_bench_mini/2d1e487f, terminal_bench_2_1/0af2ba2c): claude-opus-4.8, claude-fable-5, gpt-5.6-sol

ignore rst/transcripts (exported transcript text stays local)

rst: clause labels for every task transcript (14,957) and the 11 structure columns

README: layout, columns, provenance

features: drop 21 CSVs from crashed/partial logs kept beside the successful run

features: drop 9 empty intercode CSVs from interrupted-start stub logs

features: gpqa at data c199498b

features: cybench at data c199498b

features: aime2025 at data c199498b

features: arc_challenge at data c199498b

features: hle at data c199498b

features: aime2024 at data c199498b

features: apex_shortlist at data c199498b

features: apex_2025 at data c199498b

features: aime2026 at data c199498b

features: swe_bench_mini at data c199498b

features: intercode at data c199498b

features: terminal_bench_2_1 at data c199498b

features: mmlu_pro_math at data c199498b

features: agent_bench_os at data c199498b

initial commit