Weeks 0, 1 and 2 go under

57.4% over 760 games since 2017, under in all nine seasons and profitable in eight of them. The market is anchored: weeks 0, 1 and 2 carry the season's HIGHEST totals (55.5, 55.7, 55.1) and produce its LOWEST scoring (53.7, 53.9, 52.9), and each year the number lands within about 0.7 of last season's average. The same teams are priced 1.19 above their own week 4+ games while scoring 1.53 less. It is a mean shift and not a tail — the rule is 62.2% in one-score games, while blowouts go over — and it dies cleanly at week 3 (49.5%).

This is a statement about the market, not a forecast of the game, and it is graded on its own ledger — it never enters the model’s record. The two can disagree, and when they do the trend has the better history on these games.

The 46% line. A trend like this one can stop working — the market only has to move its opening number a point and a half to erase it, which is what happened to every early -season edge that has been published — so the condition for dropping it is set in advance rather than argued about later. If it is really still winning at its long-run rate, a season lands under 50% about one year in ten, but under 46% only about one year in sixty. So a season in the low fifties is ordinary variance and settles nothing, while a season below 46% is real evidence the market has moved and this page stops publishing it.