17. Atlanta's margin of victory over Dallas at home last night. Final ATL 86, DAL 69. The Dream beat the Wings by 17 points in a game that was effectively decided by halftime. The picks file said take DAL plus 5.5. We took the half-stake brake per the model-versus-agent conflict framework rule I wrote about yesterday morning. The half stake lost.
8. Indiana's margin of victory over Golden State at home last night. Final IND 90, GSV 82. The Fever covered the 6.5 with room to spare. The half-stake call on GSV plus 5.1 lost. The framework brake (one-of-three architectural confirmation) applied here too because the agent had zero quarter edges and no SGP correlation. Same half-stake call. Same loss.
Two half-stake losses on the slate. Minus 1 unit. The framework brake reduced what would have been a minus 2 unit slate if I had sized full stake. The brake worked. The bets did not.
The honest read on the bigger lesson.
The model-versus-agent conflict was the key call yesterday. DAL at ATL was the first explicit conflict of the season. I wrote that the picks file uses 2026 ratings and Dallas in 2026 is a better team than Dallas in 2025. The model said take DAL on the road. The agent said ATL dominates Q1, Q3, and Q4 by 3-plus net points each. Both readings were defensible inside their input data. I sized half stake to split the difference per the framework rule.
The agent was right.
Atlanta won by 17 exactly the way the quarter splits predicted. ATL Q3 was the killer (the agent said ATL Q3 plus 3.0 HIGH; the actual Q3 margin was probably bigger from the play-by-play I watched, but I do not have the official box score quarter splits in front of me to cite). The Dream are not a team you take Dallas on the road against. The 2025 net rating story is still the right story; the 2026 hot start for Dallas is variance against weaker teams (Washington was missing key players in the 92-69 game; Phoenix lost to Toronto on the same Tuesday they lost to Dallas).
The framework update going forward. When the picks file and the agent disagree on a HIGH pick AND the agent has THREE OR MORE HIGH-confidence quarter edges all favoring the same team, that is a PASS, not a half stake. The agent has three HIGH edges on the side of the team the picks file is not picking. That is a meaningful conflict, not a marginal one. Pass.
This is the third framework update of the season:
1. Half stake on partial architectural confirmation (one of three conditions met). In effect since May 18. 2. Vegas Pattern flag re-scoped to "noisy injury picture only," not "any Liberty injuries." Updated May 21 after GSV blew out NYL with two named absences. 3. PASS on model-versus-agent conflict when agent has three-plus HIGH quarter signals. New rule, effective tonight.
The framework should produce a higher win rate going forward as the architecture brakes get more sophisticated. The unit return per pick goes down because we are betting less often and at smaller sizes; the variance of the bankroll narrows. The framework optimizes for stable bankroll growth over time, not for maximizing the next slate.
Cumulative through May 22.
Spread STRONG (editorial bets): 10-9 (53 percent). Two losses last night drag the win rate down to barely above breakeven for the first time in three weeks. Total STRONG (grader): 12-10 (55 percent). Last night had a CON at SEA UNDER (we passed) and DAL at ATL UNDER (we passed) both hit, adding two wins to the totals ledger we did not size. Combined STRONG: 22-19 (54 percent). Still profitable but the spread side is the part that is suffering.
The conviction architecture is still 5-for-6. The half-stake partial-architecture profile is now 1-4 across five applications. Variance has been brutal on the half-stake calls. The win rate at 20 percent is well below the 50-55 percent the framework assumed when I set the half-stake size. Either the framework assumption is wrong or the small-sample variance is brutal. Six more applications will tell.
Tonight is small. Two games on the board. The bot's published picks file is empty (n_picks=0). No HIGH-confidence calls.
Portland at Toronto. Game card has STRONG LEAN POR plus 5.0, edge plus 7.0. Tempo home opener of the homestand. Allemand is Out. Frontcourt still thin. The matchup is closer than the line is pricing, which is the model's reading.
The framework call. With the picks file empty and only the game card flagging STRONG LEAN, we cannot apply the full architectural framework. The agent has zero quarter edges for the Tempo matchup (same one-of-three profile as the recent half-stake calls). Conservatively, this is a PASS on the spread. The STRONG UNDER 175.8 with edge minus 12.88 is the cleanest signal in the matchup. Both teams are short-handed (TOR missing four rotation players, POR missing Samuelson). The total signal is real.
Half stake on UNDER 175.8 in the Tempo home opener. PASS the spread. The framework is now consistent: STRONG UNDER signals with structural confirmation (both teams short-handed, slow projected pace) get half stake; spreads without architectural confirmation get PASS.
Minnesota at Chicago. Game card has STRONG LEAN MIN minus 2.8, edge plus 5.9. Chicago is absolutely decimated tonight. Westbeld, Stevens, Carrington, Vandersloot, and Jackson all listed Out per the live injury feed. That is the team's five most important rotation players outside of the starting frontcourt. Skylar Diggins (their best player) is in but the team around her is missing.
The model has Minnesota as an 8.7-point road favorite at neutral. The market has Minnesota as a 2.8-point road favorite. The edge is significant. The injury picture on Chicago is the calibration story (the model accounts for known absences; the market apparently does not fully). Half stake per discipline; we cannot full stake without the architectural confirmation but the matchup is closer to one-of-three-and-a-half conditions given the CHI injury context.
Half stake on MIN minus 2.8 at Chicago. The total is PASS-labeled on the card (edge minus 3.6, below threshold).
Tonight's recommended action.
HALF STAKE: UNDER 175.8 (POR at TOR). Cleanest signal of the night.
HALF STAKE: MIN minus 2.8 at Chicago. CHI is decimated; line is wrong.
PASS: POR at TOR spread (no architectural confirmation, Tempo home opener with injury uncertainty).
PASS: MIN at CHI total (PASS-labeled).
Two half stakes. Same exposure as yesterday. The framework keeps producing reads even when the variance is harsh.
The discipline tradeoff. If both half stakes lose tonight, the spread ledger drops to 10-10 and the season-long edge is gone. If both hit, the ledger moves to 12-9 and we are back in the comfortable above-breakeven territory. The slate variance is the variance. The framework is the framework.
Tempo home opener at 6 PM ET. Jordan has the Tempo Report this afternoon. Talk tomorrow morning.
[ End Report ]
Share This