What is a Predictive Accuracy Score?
A simple way to say how far a simulation can be trusted, using real experiments as the yardstick.
The idea
Suppose a simulation predicts a material’s strength, and ten published experiments measured the same property. For each pair, work out how close the prediction was: one minus the error as a fraction of the experimental value. Average those scores and multiply by 100. That is the Predictive Accuracy Score, or PAS.
Why we also report the spread
An average can hide a bad miss. So we also report the spread of the individual scores. A high PAS with a small spread means the model is consistently close. A high PAS with a large spread means it is sometimes very wrong.
Thresholds
Our framework aims for a PAS of at least 80% with a spread of at most 15% across ten or more datasets. Safety-critical uses aim higher: 85% and 12% across twenty or more.
Try it
Use the calculator on the home page with your own numbers. It runs in your browser and is indicative only.