One-day-ahead regression study
Dataset: UCI Bike Sharing daily counts, CC BY 4.0
Selected on validation: previous_day
Final observations: 110
Selected final MAE: 926.573 requests per day
Mean residual: -46.736 requests; positive means underprediction
Largest absolute residual: 4470.000 requests on 2012-10-31
Predictions remain real-valued, with no unannounced rounding or clipping.
Final results do not justify changing the frozen selection. The public fixture is an evaluation rehearsal, not a hidden benchmark.

Actual and predicted daily requests for 110 final-test dates. Final MAE is 926.573 requests/day. Exact values are in the selected-predictions table; a close line match does not establish reliable future performance.

Residuals are actual minus predicted; above zero means underprediction. Mean residual is -46.736 requests. Largest absolute residual is 4470.000 on 2012-10-31. Inspect the full table for changing error direction, runs, and isolated errors; errors need not be independent.