Anyone who has been burned by a tool that projected with more confidence than it had earned.
Why you should believe any of this
How do I know the forecast is not just confident nonsense?
Because it grades itself in public, and it has already failed twice in ways you can read about.
Walk-forward validation, shipped as a screen
The model is placed at points in your past, given only what was knowable then, and scored against what actually happened. Most forecasting products never do this. The ones that do rarely show you.
Coverage and sharpness are reported separately
A p05–p95 band should contain reality 90% of the time. More means it is hedging; less means it is lying. And narrow bands are only a virtue while coverage holds — anyone can be perfectly calibrated by predicting 'somewhere between bankrupt and rich.'
The page can deliver bad news about the model
It already has. It caught a synthetic-data defect, then a persistent optimistic lean that turned out to be credit card payments being misclassified as internal transfers. Both are written up in the roadmap rather than quietly patched.
Also built for