Skip to main content
Search

# Prediction Card

1 related articles

Subscribe to updates
  1. arXiv

    Paired forecasts across 336 prospective states leave explanation credit unconfirmed, while structured cards cut Tox21 repeat drift by about 60 percent

    The work introduces a predictive-credit protocol that compares description, matched explanation, and donor explanation under paired forecasts sharing an intervention, forecaster, and outcome across 336 prospective states (controlled learning, 12 Tox21 endpoints, 24 OpenML tasks); the frozen v5 decision was inconclusive and Tox21 and OpenML returned confirmation_no_go, matched point-accuracy gains remained unconfirmed, yet structured cards reduced Tox21 repeat drift by 64.5 and 59.1 percent and a researcher-authored mechanism positive control lowered point MAE by 2.60 percentage points.

You’re all caught up