Research preview
Satcove Accuracy Index
No public Accuracy Index has been released yet. This page is excluded from search results and does not publish model rankings, accuracy percentages, downloadable datasets or sample-size claims.
Required before publication
Reproducible corpus
The evaluated prompts, category definitions, inclusion rules and model run dates must be versioned and publishable. Private user conversations will not be used as a public benchmark corpus.
Panel integrity
Each run must record the provider and model actually executed, including failures and fallbacks. Several labels backed by one provider will not count as independent votes.
Separate measurements
Panel agreement, factual correctness, source coverage and freshness will be reported separately. Agreement alone will never be presented as accuracy or probability of truth.
Auditable results
The publication must include denominators, uncertainty intervals, scoring rubrics, evaluator conflicts, failed runs and machine-readable data before any leaderboard claim is made.
Publication status
Publication remains blocked until every gate above is met and the evaluation can be reproduced outside Satcove. Any future release will carry a dated methodology version and a changelog.
Contact the research team