-
Add Qwen explicit-format and checker-retry development results; preserve v1
Publish verified benchmark results and accurately report incomplete runs. Preserve dataset files, original benchmark results and license wording. -
Add Qwen results on 204 oracle protocol compliance cases
Publish verified benchmark results and accurately report incomplete runs. Preserve dataset files, original benchmark results and license wording.
-
Add verified oracle worker pilot and clarify historical benchmark scope
Publish verified benchmark results and accurately report incomplete runs. Preserve dataset files, original benchmark results and license wording. -
Add Opus benchmark and balanced quality, speed and token ranking
Publish verified benchmark results and accurately report incomplete runs. Preserve dataset files, original benchmark results and license wording. -
Add Astra xhigh, token usage and response time to the benchmark ranking
Publish verified benchmark results and accurately report incomplete runs. Preserve dataset files, original benchmark results and license wording. -
Rank model benchmarks by F1 and add Sonnet 5.5
Publish verified benchmark results and accurately report incomplete runs. Preserve dataset files, original benchmark results and license wording. -
Add Claude Haiku 5.5 benchmark result
Publish verified benchmark results and accurately report incomplete runs. Preserve dataset files, original benchmark results and license wording. -
Add four Qwen Ambassador benchmark results
Publish verified benchmark results and accurately report incomplete runs. Preserve dataset files, original benchmark results and license wording.
-
Add current model benchmarks and GPT-6.1 Sol results
Publish verified benchmark results and accurately report incomplete runs. Preserve dataset files, original benchmark results and license wording. -
Shorten dataset card and document original-data model benchmarks
-
Refresh oracle corpus and add 2026-10-13 benchmark
Update public oracle archive and audit; add dated temporal splits while preserving the original benchmark byte-for-byte.
-
Publish IMD oracle outcomes snapshot from 2026-10-08
Add 7,284-request provenance archive, 5,941 benchmark examples across train/validation/test, dataset documentation, audit metadata, and checksums. Source license remains unknown.