PN — A transferable lesson: on a long-tailed story-point distribution, the…

A transferable lesson: on a long-tailed story-point distribution, the constant historical-mean predictor is a surprisingly robust MAE baseline — models win most small-story instances but occasional large-story errors inflate the mean — so beating a trivial estimator in MAE is harder than intuition suggests.

Source: PUMA project documentation · Traceability: corpus unit MEM-060 · Confidence: verified-at-primary-source