5.5. The market spread on Phoenix at Atlanta yesterday. Final ATL 82, PHO 80. The Mercury lost the game but covered the spread by 3.5 points. The bot published this as its only HIGH-confidence pick of the day. We passed on it per the discipline framework (no conviction architecture, MODEST signals across the board, no architectural confirmation). The pass cost us 0.91 units of profit.
The pattern is becoming familiar.
POR plus 11.3 against the Liberty two Mondays ago — passed per Vegas Pattern, POR won by 18. POR plus 11.8 against the Liberty the prior Monday — passed per Vegas Pattern, POR won by 18. GSV plus 7.5 against the Liberty on May 21 — passed per Vegas Pattern, GSV won by 17. PHO plus 5.5 at Atlanta yesterday — passed per no-conviction discipline, PHO covered by 3.5.
Four missed wins on published HIGH picks in the last 17 days. The cumulative miss is approximately 3.6 units of would-have-been profit.
The framework's case for itself. The four missed wins are offset by a 11-9 record on the editorial-sized spread STRONG bets (55 percent) and a 5-1 record on conviction architecture calls (83 percent). Per the grader, including all bot-labeled STRONG LEAN cards (which is closer to "what the model would have bet at full size"), the spread STRONG ledger is 14-8 (63.6 percent) with +21.49 percent ROI. The discipline keeps the bankroll above breakeven; the misses do not break the math.
The framework's case against itself. If the editorial discipline had matched the model's published picks at half stake on the four missed games, we would be at approximately plus 1.8 units of additional profit (4 wins at half stake at -110, 4 x 0.45 unit). That is not a small amount over a long season. The discipline is filtering for the wrong things.
Two ways to read this honestly.
The pessimist read. The framework rules I have been adding (Vegas Pattern, partial architectural confirmation, model-versus-agent conflict) are over-engineered. The bot's published picks are above 60 percent on their own. By layering filters on top of them, we are turning a profitable system into a barely-profitable one. The simpler approach is to bet the bot's HIGH picks at half stake by default and reserve full stake for conviction architecture.
The optimist read. The framework is a sample-size game. Four missed wins in 17 days is variance, not signal. The Vegas Pattern was right twice before it was wrong (and even the recent GSV/DAL hits were not as overwhelming as the line implied). The conviction architecture went 5-for-6 by design, not by luck. The framework is doing its job; the slate variance is hurting us at the margins where the framework cannot be sure.
I lean optimist still, but the lean is weaker than it was last week. Two more weeks of data will tell. If the missed-win count gets to 6 or 7 in the next stretch, I will revisit the framework filters with a default of "follow the bot's HIGH at half stake unless a conviction signal fires for full stake."
The cumulative through May 24.
Spread STRONG (editorial bets): 11-9 (55 percent). No new bets last night. Total STRONG (editorial bets): 8-7 (53 percent). No new bets last night. Combined STRONG (editorial): 19-16 (54 percent).
Per the grader (all STRONG LEAN cards including ones we did not bet): spread 14-8 (63.6 percent, +21.5 percent ROI). Total 13-12 (52 percent, -0.7 percent ROI). The bot's model is significantly more profitable than the editorial-sized version. That gap is the cost of the discipline framework.
Tonight is small. Two games.
The bot's picks file shows zero HIGH-confidence calls tonight (n_picks=0). The next-best signals come from the game card labels.
Connecticut at Golden State. Card label: STRONG LEAN CON plus 12.4, edge plus 6.95. The picks file does not publish this as HIGH because of the calibration confidence.
The framework conflict. The agent has THREE HIGH-confidence quarter signals all favoring GSV (Q1 plus 3.50, Q2 plus 2.80, Q3 plus 3.90). The agent's SGP correlation reads "GSV dominant game." That is the exact pattern that triggered the new framework rule on May 23 after the DAL at Atlanta loss: PASS on 3-plus HIGH agent signals contradicting the picks file. We tested that rule once (DAL at ATL was the conflict that prompted the rule; we sized half stake before the rule existed, and that bet lost). Tonight is the second application of the rule.
PASS: CON plus 12.4 at GSV. Per the model-versus-agent conflict rule. If GSV covers, the rule is 2-for-2 and stays. If CON covers, the rule needs another test.
Portland at New York. Card label: PASS spread (edge plus 12.91 but the expansion flag on Portland triggers an auto-pass). Picks file does not publish a HIGH pick here either.
The under signal is strong. STRONG LEAN UNDER 177.2 with edge minus 13.4. The Liberty are still missing Satou Sabally and Sabrina Ionescu plus Leonie Fiebich and Marine Fauthoux. Portland is missing Karlie Samuelson. Both teams are thinned. The projected pace is slow. The matchup screens for a low-scoring game.
The Vegas Pattern question. New York has been mispriced by the market for three straight weeks against teams that the model said would cover. The pattern was wrong twice in a row recently (GSV blew them out by 17; DAL by 15). The reframing rule from Issue 19 says the Vegas Pattern flag applies to GENUINELY NOISY injury pictures, not CLEAN-BUT-SEVERE ones. Tonight's NYL absentee list is the latter (four named players Out, full lineup otherwise available). The flag does not fire.
Half stake on UNDER 177.2 in POR at NYL. The signal is real, the structural conditions are present (both teams thinned, slow projected pace), and the Vegas Pattern flag does not apply.
Tonight's recommended action.
HALF STAKE: UNDER 177.2 (POR at NYL). Slow on slow, both teams thinned, no Vegas Pattern caveat.
PASS: CON at GSV spread (3-plus HIGH agent conflict rule).
PASS: POR at NYL spread (expansion flag on the card, no published bet).
PASS: CON at GSV total (PASS-labeled, no edge).
One bet tonight. The framework keeps producing reads even on a slate with no published HIGH picks.
Tempo plays tomorrow (Tuesday) at home against Phoenix in the second game of the homestand. Allemand's status is the rotation question; if she is back the Tempo are favored. Jordan will have the Tempo Report tomorrow morning. The homestand finishes Thursday (game three TBD per the broadcast schedule).
Talk tomorrow morning. The framework holds, sample-size questions and all.
[ End Report ]
Share This