Touchstone: rigor ratings for AI benchmarks

Skip to content

pact/touchstone

Ratings
Standards
How it works
GitHub

.skel-page{display:none}Touchstone needs JavaScript to show the ratings.The data and scoring rules are open: see the repository.

Loading

A PACT project. Scores are a deterministic function of the verdicts. Positive verdicts cite pinned evidence; the verdicts themselves are judgements.

Keyboard shortcuts

/: Search benchmarks
g then r: Go to ratings
g then s: Go to standards
g then m: Go to how it works
Esc: Back to the ratings from a benchmark or comparison
?: Show this list