33. That is the margin Phoenix beat Las Vegas by last night. We took PHO +9. The Mercury did not need the points. They won the game outright, 99 to 66, in front of the Vegas home crowd that the market had spent two days pricing as the favorite.
The Vegas Pattern thesis was correct. The 2025 priors that drove our model were right. The market that priced the Aces as 8-point favorites was wrong. We called it. The bet won. The bet won at half stake.
That is the part I want to write about today.
We had three STRONG LEANS on yesterday's slate. Two of them were full stake. One of them, the one with the most edge by the model (+10.85 points), was sized down to half because we identified the pattern and flagged the prior risk. The full-stake picks were Portland +5.5 and Dallas-Indiana UNDER 178.2. Both lost. Portland lost outright by 15. The Dallas-Indiana total went 211, thirty-three points over the line. Both were brutal misses.
The half-stake pick covered by twenty-four points.
This is the bankroll math. If we had reversed the sizing, betting full stake on Phoenix and half stake on the others, we would have grossed +0.91 units on PHO and lost 0.5 units on each of the others, for a net of -0.09 units. The actual sizing was +0.45 on PHO and -1.0 on each loss, for -1.55 units. The discipline cost us about a unit and a half on the night.
Here is the question. Was the discipline still right?
Yes. The Vegas pick had a known structural risk. The 2025 prior was probably underpricing a Vegas roster that loaded up in the offseason. Either the model was right that the loaded Aces would not gel quickly, in which case Phoenix covers easily, OR the model was wrong because the new roster integrates immediately, in which case Vegas covers easily. The probability distribution on the Vegas pick was wider than on the other picks. Wider distributions get smaller bets. That is how Kelly sizing works. It is how it should work.
The win felt big because it was a 33-point margin. But on a normal random pull from the model's distribution for that game, the average margin was probably closer to 2 points in either direction. We got the high end of the distribution. That is variance, not skill. The reverse outcome was just as likely.
The full-stake losses are a different conversation.
Portland is an expansion team with no 2025 data. Our model defaulted them to league-average ratings. Chicago is in year one of a rebuild and was missing both Cardoso (rest) and Vandersloot (back). League average against a banged-up rebuilding team should win at home. It did not. Portland was not league average last night. They were worse. The lesson: expansion priors are too optimistic. We will adjust.
Dallas-Indiana UNDER had the strongest model edge on the slate (-10.82). The total was 178.2. The model said 167.4. The actual total was 211. The number to look at is not the total. It is the pace. Indiana ran 84 possessions. Dallas ran 86. Their season averages last year were 80 and 82. Both teams played significantly faster than expected. There is no model in the world that predicts that pace shift on opening night.
Both losses had a specific story. The PHO win had a specific story. The model is volatile in early season because the 2025 priors are doing too much work. By Memorial Day we will have ten 2026 games per team and the priors will adjust. Until then, every pick should be sized smaller than the model suggests.
Tonight's slate. Four games. Three leans. Walking through them with explicit sizing.
SEA at CON, STRONG LEAN SEA -1.6. Edge +8.25. The Storm are road favorites by 1.6 in the model and the market has them as a small underdog. Sue Bird is in the building tonight as part of the Magbegor injury coverage. Half stake. The Connecticut Sun are a known unknown right now and the Magbegor absence makes the Storm's number harder to trust.
LVA at LAS, STRONG LEAN UNDER 176.8. Edge -8.50. Both teams play deliberate offense. LA is missing Plum (illness, listed questionable but already ruled out by beat reporters). Vegas is on a back-to-back after the loss to Phoenix. Two teams with rest issues, missing pieces, in a slow game environment. Full stake. This is the cleanest pick on the board.
LVA at LAS, MODEST LEAN LVA -2.5. Edge +2.95. The model has Vegas favored on the road by 2.5. The market has them at 0. Same Vegas team that just got demolished by 33 last night, now playing a back-to-back on the road. The model does not know about the back-to-back. Pass on this one. Skip.
PHO at GSV, MODEST LEAN PHO +1.7. Edge +4.50. Phoenix is on a back-to-back too. They just dropped 99 against Vegas, which is their season high. The model does not know about either fact. Skip.
So tonight: one full stake (UNDER LVA-LAS 176.8), one half stake (SEA -1.6), two skips (both back-to-back games where the model is blind to fatigue). That is conservative and that is correct.
Adding back-to-back to the flagged-priors list. The model treats every game the same regardless of rest. Day-to-day NBA games have a documented fatigue effect of about 2-3 points against the team on the back end. WNBA games are less studied but the effect should be similar. Until the model accounts for this, treat any back-to-back team's bet as a full tier downgrade. STRONG becomes MODEST. MODEST becomes PASS. SKIP stays SKIP.
The picks scoreboard updates tomorrow morning. Cumulative through two nights: 2 wins, 5 losses. The Vegas pattern was the win that mattered. The bankroll discipline was right. The math is brutal. Both can be true.
[ End Report ]
Share This