A few notes on the results:
When publishing the picks, I had inadvertently scaled down the bet size by 50%. I have a parameter in my code that allows for Kelly fractional betting, and I had that parameter set at 0.50. In practice, most gamblers/investors only bet a pre-determined fraction of what the Kelly strategy recommends. In effect, it's a hedge against your model probabilities being incorrect. So, while the original post indicated we were betting 7.8% of our bankroll, the actual recommended bet size was almost twice that, at 14.6%, as reflected above.
For each week's picks, I deliberately avoid publishing the point spread. The lines move as the week progresses, and the model was calibrated against the closing line. Success of the betting strategy is judged on its ability to beat that closing line. My source for that closing line is sportsdatabase.com, if you want to check my work.
Picks went 1-2 against the spread last week. One of those unsuccessful bets was on the Patriots (to cover as a favorite against the Jets). According to sportsdatabase, the closing line was 7.5 points, and the Pats won by 7. Most books seem to have had the line at 7 points, which would have made for a push, but I'm sticking with the sportsdatabase number for consistency. When judging the success of a betting strategy, the temptation to cherry pick is real.