27. The Valkyries' margin of victory at home against Connecticut on Monday. Final GSV 97, CON 70. The bot's published picks file had CON plus 12.4 as a STRONG LEAN. We passed per the model-versus-agent conflict rule (the agent had three HIGH-confidence quarter signals all favoring GSV; the picks file picked CON). The pass was right. GSV covered the line by 14.5 in the agent's predicted direction.
That is the third application of the conflict rule. The first was DAL at ATL on May 22 (we sized half stake before the rule existed; ATL won by 17 in the direction the agent predicted, validating what would have become the rule). The second was tonight's CON at GSV (PASS per the rule; GSV won by 27 in the direction the agent predicted). Two for two on the rule when applied. The framework filter is producing the right read when the agent and picks file disagree by enough.
22. The margin by which POR at NYL cleared the UNDER 177.2 we sized at half stake. Final POR 81, NYL 74. Game went 155. Cleared by 22.2. The Vegas Pattern reframing held — clean-but-severe NYL injury picture is not the same as the noisy injury picture the original Vegas Pattern flag was meant to filter. Half stake on the under was the right call. Plus 0.45 units.
Net Monday slate: plus 0.45 units. Tuesday was a no-games rest day. The cumulative editorial ledger through May 26 sits at spread STRONG 11-9 (55 percent), total STRONG 9-7 (56 percent), conviction architecture 5-1 (83 percent).
Now the more interesting frame question.
Yesterday I wrote that the framework filters have cost approximately 3.6 units in missed wins on published HIGH picks across the last 17 days. The optimist read was that the framework discipline keeps the bankroll above breakeven. The pessimist read was that the filters are over-engineered and the bot's HIGH picks at half stake by default would have produced more profit.
The framework update I floated yesterday: "default to half stake on the bot's HIGH picks unless a conviction signal fires for full stake or a hard discipline brake (model-versus-agent conflict, Vegas Pattern when noisy) fires for PASS."
Tonight is the first slate where the new default rule applies cleanly. The bot's picks file has TWO HIGH-confidence published picks.
Toronto at Chicago. TOR minus 5.5. Edge plus 6.95. Architecture has one of three conditions present (model has road team favored). Agent has no quarter signals. No discipline brake fires. Half stake by default.
Atlanta at Minnesota. ATL plus 2.6. Edge plus 6.4. Architecture has one of three conditions present (model has road team favored). Agent shows MIN Q3 plus 1.9 MEDIUM (a small home-team quarter edge, not 3-plus HIGH signals). No hard discipline brake fires. Half stake by default.
Both at half stake. Per the new default rule.
The case for following the rule strictly tonight. The bot's HIGH picks are running 14-9 (61 percent) on the season per the grader. At half stake at minus 110, breakeven is 52.4 percent. Half stake at 61 percent produces approximately plus 0.18 units per pick on expected value. Two picks tonight equals plus 0.36 EV.
The case against. Both architectural confirmations are weak (one of three each). The TOR matchup has a Tempo team that lost at home by 19 four days ago. The ATL matchup has Minnesota at home, where they have been very good even without Collier (something like 7-1 at home this season per my count). The expected value is real but the variance per game is high.
I follow the rule. Half stake on both. The cumulative variance washes out over a 44-game season.
Tonight's recommended action.
HALF STAKE: TOR minus 5.5 at Chicago. Published HIGH pick. One of three architecture. Default half stake.
HALF STAKE: ATL plus 2.6 at Minnesota. Published HIGH pick. One of three architecture. Default half stake.
PASS: POR at CON (no game card generated tonight per the bot).
PASS: NYL at PHO (no game card generated).
PASS: WAS at SEA spread (MODEST tier per the card). PASS the total too (no edge).
Two bets at half stake. Exposure approximately 0.9 units. Expected value approximately plus 0.4 units if the model's edge holds at 61 percent.
The third application of the half-stake default rule will tell us whether the new framing is the right one. If both hit, the spread editorial ledger jumps to 13-9 (59 percent). If both lose, it drops to 11-11 (50 percent, dead even). If they split, we are 12-10 (55 percent, status quo).
Talk tomorrow morning. Jordan has the Tempo Report this afternoon on the TOR at CHI matchup specifically.
[ End Report ]
Share This