What has been checked, what has not, and what stays research-private.
Forecast skill is not yet validated against observed outcomes. What is published today is a scorecard of per-episode detection counts, together with an explicit statement of what those numbers cannot support.
The scorecard reports detection counts for five historical episodes at an alarm threshold of 0.500 on the class severity score:
Detection counts for five episodes do not establish probability calibration, skill against a climatology baseline, or a false-alarm rate. Until validation against observed outcomes ships, treat every severity value as decision support only.
Model code, dataset collection procedures, training and benchmarking are research-private. Only results and outputs are public, on this site and in the repository. For research collaboration or licensing enquiries, use the contact page.
Loading the interactive HazardNet application…