Extended Data Fig. 2. Distribution of normalised errors from both anhydrate and hydrate validation sets.

$\left(\mathrm{\Delta }\mathrm{F}-\mathrm{\Delta }\hat{\mathrm{F}}\right)/{\sigma }_{\mathrm{\Delta }\mathrm{F}}$ is expected to have a mean of 0 and a standard deviation of 1 if $\mathrm{\Delta }\mathrm{F}-\mathrm{\Delta }\hat{\mathrm{F}}$ is normally distributed.