Team projected talent (e.g. WAR etc.) figures in from the beginning and throughout the first half of the season. These act as buffers for freaky records early. For example TB’s projection vs the Sox despite the early season disparity in records.projections are too reactionary to recent play. they're just now starting to balance out.
The algorithm(s) lean heavily on projected talent level early and record is secondary. This gradually moves toward record > projected talent as the season progresses and real data (“True Positives”) are generated.
I find theses projections fun and they have a reasonable degree of fidelity.