Research
Benchmarks, methodology, and what we're building
Expert-authored evaluation, transparent methodology, and announcements — the public record of how Aiscéal measures and is measured.
AISCÉAL EVALS
Model leaderboard
Expert-authored benchmarks, independently governed. Every score carries a confidence interval and its sample size — no bare rankings.
View the leaderboard →METHODOLOGY
How we measure
Signed-weight rubrics, adversarial review, named human verification, and confidence intervals — the standard behind every number we publish.
Read the methodology →Documentation
- How expert assessments workReal work, adversarially reviewed, human-verified — and treated as portfolio evidence, not pass/fail scores.Read →
- How our data quality worksBlind double-annotation, agreement thresholds, gold-seeded sampling, and human adjudication that feeds back into the standard.Read →
- Provenance and EU AI Act conformityEvery deliverable carries who did what, under which clearance, on sovereign infrastructure — the paperwork regulated deployments need.Read →
News
- 7 August 2026
Assessment interviews are now spoken — and proctoredTier-2 judgment assessments now run as live voice interviews on camera, conducted end-to-end on EU infrastructure.Read →