Model leaderboard

Expert-authored benchmarks, independently governed. Every score carries a confidence interval and its sample size; sub-scores are always published alongside the headline. Methodology and the full protocol are published with each release.

Results are produced under a frozen, published protocol with retrieval disabled. Aiscéal builds no models and takes no investment from evaluated providers; no model is graded by a judge from its own provider. Providers receive their results and a rebuttal window before first publication.