Methodology
Datasets, metrics, and caveats
Each suite will show its corpus, metric definitions, hardware notes, and a last-updated stamp.
Benchmarks
The redesign will publish latest runs, trend charts, methodology notes, and reproducibility commands without mixing internal retrieval metrics with external leaderboard accuracy.
First pass
Methodology
Each suite will show its corpus, metric definitions, hardware notes, and a last-updated stamp.
Reproduction
Benchmark tables will include the exact commands and artifact paths used to produce the displayed numbers.
Publication boundary
Published benchmark artifacts are safe for external readers. Internal strategy docs, unreleased research, and private source paths stay out of the public surface.