First! ;-)

I like this thought! Clearly, winning probability increases (monotonically, I’d wager) with spread (from + to -). However, the relation is not linear; your fitted line implies that a team favored by 17 or more would have >100% winning chances. I’ve recently been trying to model this relationship out, and an ERF (a linear transformation of the Gaussian distribution CDF) seems to work better. If you use the point spread as the x-axis, the ERF implementation of Google sheets approximates historical win prob quite well:

win_prob = (1+erf(-spread/17))/2

For instance, if you put in +3, you get win_prob = 40.1%; fairly close to 41.5% historical. With -10, you get 79.7% vs 79.3% actual. There’s not enough data in NFL history to get a really smooth curve, especially for double-digit spreads (and sample sizes get really small for spreads of 15+), so there’s a fairly wide margin of error regardless of model. But for such a simple function, ERF does pretty well!