promptdojo_

Regression metrics and residuals — step 5 of 7

This report brags MAE = 2.8 minutes — after silently DELETING every row with an error over 20 minutes. The deleted row is a 90-minute delivery predicted at 45: possibly a data error, possibly the product's worst failure, definitely not something to hide. Fix the report to compute MAE over ALL rows and print the suspicious rows under "investigate:" instead of deleting them.

The break is on line 3 — but read the whole snippet first.

full-screen editor opens — close anytime to keep reading.