23. The number of points Dallas beat Seattle by last night. Final 79-56. The architecture's first post-exclusion test went the wrong way at the reduced sizing.
The framework call yesterday was half stake on SEA plus 12.5 per the Dallas exclusion rule that we set Friday. Half stake because the rule said size down on Dallas applications. Not full because the architecture's Dallas record was 2-2 going in. Not pass because the alignment math was the cleanest of the season.
The result resolved the rule's first live test in the direction of "the rule was right to size down." Half stake at 0.5 units of risk produced a 0.5-unit loss after juice. Full stake would have been 1 unit risked and 1 unit lost. The rule saved exactly the variance it was designed to save.
The architecture's standing.
Non-Dallas applications: 5-for-5 (one hundred percent). Unchanged.
Dallas applications: 2-for-5 (40 percent). Now three straight losses. WAS by 23 on May 18. LVA by 8 on May 28. SEA by 23 last night. Three of the last five architecture applications have been at Dallas. Three losses. The pattern is no longer noise.
The framework needs to address Dallas specifically again.
The original exclusion rule said "size down to half stake or pass." We sized down to half last night. The half stake lost. The next question is whether the next Dallas application gets sized to quarter stake, sized to zero, or sized to half stake again with the rule unchanged.
My read after one live test: the rule held. Half stake on a Dallas application that lost is 0.5 units of damage, which is the same as the modal half-stake-tier loss the rest of the season has shown. The Dallas-specific variance is not bigger than the half-stake-tier variance generally. The exclusion rule treats Dallas the same way the half-stake tier treats every weaker-conviction call. That is the right size.
The next Dallas application gets half stake again. Three losses in five does not flip the rule to "pass everything at Dallas" because the architectural alignment math, when it fires, has produced two wins in five Dallas tries — which is the same hit rate as the half-stake tier overall (around 40 percent). The rule is functioning as a half-stake tier override of the architecture's normal full-stake sizing, and that is the right shape.
The bot's headline ledger from yesterday.
Spread STRONG: 12-12 (50 percent). One loss adds to the line. Functionally a coin flip. Total STRONG: 9-9 (50 percent). No change yesterday (SEA-DAL total was PASS on both sides, MIN-PHO total was MODEST so not in the STRONG ledger).
Per the grader on all bot-labeled STRONG cards including ones we did not size: spread 16-14 (53 percent), total 15-16 (48 percent). The bot's record is regressing back toward the backtested baseline of 57.9 percent on spread STRONG. Live performance for May was above baseline for the first three weeks and has been at-or-below baseline for the last two.
The MIN-PHO blowout last night is also worth noting for the model-agent conflict rule.
The bot card had MODEST LEAN PHO with edge minus 2.95. The agent had MIN Q3 plus 3.1 HIGH and MIN Q4 plus 1.7 medium — two agent quarter edges on the opposite side from the bot pick. That is below the formal 3-plus HIGH threshold for the model-versus-agent conflict rule. Final score Minnesota 111, Phoenix 77. Minnesota won by 34. The agent was right and the bot was wrong by 31 points beyond the spread.
That tells us the conflict rule's threshold may be too high at 3-plus HIGH. Yesterday's MIN-PHO was 1 HIGH plus 1 medium on the opposite side and the agent was correct by 31 points outside the spread. The threshold review for the conflict rule's HIGH-count needs to happen later this week. For now, the formal rule is 3-plus HIGH and we follow it.
Tonight is PASS-all across four games.
The slate has four games on the board. The bot's published picks file shows zero HIGH-confidence calls. Going through the cards by tier:
CON at ATL. The bot card has the spread as a STRONG LEAN on CON at edge plus 5.05. The agent has the opposite read at the most extreme firing of the model-versus-agent conflict rule we have ever recorded: ATL Q1 HIGH plus 5.5, ATL Q2 HIGH plus 2.5, ATL Q3 HIGH plus 5.6, ATL Q4 HIGH plus 4.1. Four HIGH signals, all on the opposite side from the bot pick, with an "ATL dominant game" SGP correlation. The conflict rule fires hard. PASS the spread. PASS the total (small edge, no STRONG label).
POR at GSV. The bot card has the spread as a STRONG LEAN on POR at edge plus 9.51. The agent is completely silent (zero quarter edges, zero player props, zero SGP correlations). The model is alone on this call. The conviction architecture does not fire here because POR is not favored at neutral (the model has GSV plus 1.11 at neutral). This is a half-stake candidate by the partial-confirmation rule. After yesterday's half-stake loss, the half-stake tier's drag on the season's ledger is the heaviest of any tier (now roughly 1-7 across the year). PASS. The tier is dragging too hard to fight a silent agent.
LVA at LAS. Spread PASS per the card. Total MODEST LEAN UNDER 176 at edge minus 7.7. MODEST UNDER without structural confirmation gets a PASS by default. Both teams are thinned (LAS missing Plum, LVA missing Evans), which is the kind of profile that could activate the structural rule — but the rule requires the STRONG UNDER label to apply, and this is MODEST. PASS.
CHI at WAS. Spread PASS. Total PASS. Both teams thinned (CHI missing 4 players including Vandersloot and Carrington, WAS missing Citron). PASS by tier.
Zero exposure tonight. Nina has the tactical piece this morning on the CON-at-ATL conflict. Jordan has the Tempo Report pulled forward by a day with the New York preview for Wednesday's matchup.
Tomorrow's preview.
Tempo at New York is the Wednesday headline. The Liberty are the league's top team. The Tempo are 5-5 coming off Saturday's home win. The architecture does not fire (Toronto is the road dog, not the road favorite). The matchup is the test of whether the 5-5 expansion arc holds against tier-one opposition.
The slate beyond Wednesday opens up. Multiple architecture candidates may fire later in the week as the schedule resumes its full pace. The non-Dallas architecture continues to look like the framework's strongest weapon at 5-for-5.
The discipline holds. Tonight PASSes. The rule earned its keep on a losing night. Talk tomorrow.
[ End Report ]
Share This