Dailies
review the day's footage, find what comes before the bad takes
Screen your own log and see what tends to come before your bad days. Nothing leaves this tab.
How it worksLoad
Paste or upload a diary export, or try the sample. Nothing leaves this tab.
The sample is 120 made up diary days. A hidden rule makes bad days far more likely after short sleep together with a skipped meal. See if the screen finds it.
nothing loaded yet
Cast
Choose what a bad day means and which factors get a part.
Load a log to cast it.
Screening
Five folds, each holding out a stretch of days the models never train on.
The network appears once screening starts.
compares every backprop gradient, weights and inputs, with a central difference
Verdict
Did anything beat simply guessing your usual rate?
Run the screen to get a verdict.
Credits
Which factors moved the risk most, and which way.
Credits roll after the screen.
What if
Set today's values and read the estimate, with its spread.
Available after the screen.