1. The number of units the framework discipline saved last night across the bot's 0-for-3 on HIGH-confidence spread calls.
The bot's published picks file had three HIGH spread calls Saturday. Indiana plus 3.5 at New York. Golden State plus 3 at Las Vegas. Washington plus 10.2 at Atlanta. All three lost.
The IND at NYL piece Saturday morning identified four data-freshness flags that argued for PASS despite the strong card label (silent agent, stale Ionescu prop, 20-point disagreement between card and picks file on the total line, no architecture firing). The four-flag PASS was correct. Indiana lost 83 to 75 outright. Indiana plus 3.5 lost by 4.5 against the spread.
A full-stake bet on IND would have lost 1 unit. A half-stake bet (the partial-confirmation default sizing) would have lost 0.5 units. The PASS held. The framework filter saved between 0.5 and 1 unit on that single game.
The other two HIGH spread losses were not editorial bets at the same flag-density. The GSV at LVA spread closed at minus 3 with the bot's HIGH on GSV at the +3 dog. Golden State lost by 5. The WAS at ATL spread closed at minus 10.2 with the bot's HIGH on Washington at the +10.2 dog. Atlanta won by 32. The Washington spread lost by 22 — the worst against-the-spread blowout the bot has produced this season.
The Washington blowout is the model-versus-agent conflict rule's threshold-review evidence.
The bot card on WAS at ATL had the spread as MODEST LEAN WAS at +10.2. The agent had ATL Q1 plus 4.3 HIGH, ATL Q2 plus 4.9 HIGH, ATL Q3 plus 2.2 medium, and an "ATL dominant game" SGP correlation. Two HIGH signals on the opposite side from the bot pick. The formal conflict rule requires three-plus HIGH signals on the opposite side. Yesterday's firing was below the threshold by one HIGH signal. The agent was correct by 22 points against the spread.
The threshold review needs to happen this week. Two HIGH signals on the opposite side plus an SGP narrative may be enough to flip the rule to PASS. The data argues for a 2-HIGH-plus-SGP threshold or a 2-HIGH-with-margin threshold. The 3-HIGH threshold is too conservative on the upside (we PASS the spread either way at MODEST tier) but is the wrong filter for the future case where the agent's two-HIGH-plus-SGP read is on the opposite side of a STRONG bot pick. The next time that scenario presents we want the rule to fire.
The STRONG UNDER record from Saturday is the other clean filter.
Both the GSV at LVA UNDER (line 169.5) and the IND at NYL UNDER (line 175.3) won. GSV-LVA went 163 combined. IND-NYL went 158 combined. The structural-confirmation rule's framework hand the discipline a half-stake-tier opportunity on either game and both would have won at 0.5 units each (plus 0.9 units combined after juice).
We did not size either because the agent was silent on both and the discipline has been PASSing silent-agent STRONG-tier reads for a week. The pattern is correct on the spread side and incorrect on the totals side this week. The structural-confirmation rule's tradeoff is real: we save spread variance by PASSing silent-agent partial-confirmation plays, and we miss totals upside by PASSing silent-agent structural-confirmation plays.
Tonight has the same shape on both sides.
Two STRONG UNDERS on the slate. CHI at TOR has the largest UNDER edge of the week (11.44 points) with both teams thinned cleanly (CHI missing four, TOR missing three including Rice). POR at LAS has a STRONG UNDER 177.1 with the edge minus 9.48. The structural-confirmation rule's injury condition is the cleanest of the season on the CHI-TOR side. The pace condition is plausibly met. Both signals point toward the half-stake-tier opportunity the rule was designed for.
One HIGH bot spread on the slate. POR at LAS is the bot's picks-file HIGH at 10.1 points of edge with the side on POR at +7.7. The architecture does not fire (LAS is favored at neutral, road team not favored). The agent is silent on quarter signals (two medium player props on Plum and Hamby fatigue, nothing else). This is the same shape as yesterday's IND at NYL.
The framework discipline argues for PASS on all three. The pattern has been six consecutive correct PASS calls on silent-agent STRONG-tier reads. The seventh PASS today extends the pattern. Sizing a half stake tonight breaks the consistency that has been the framework's actual edge this week.
The verdict.
PASS the CHI at TOR UNDER. PASS the POR at LAS spread. PASS the POR at LAS UNDER. Zero exposure tonight.
The cumulative ledger.
Spread STRONG editorial-sized: 12-13 entering tonight. Yesterday added zero (we sized nothing). The bot's record on STRONG cards across all graders is roughly 18-20 on spreads now, drifting below 50 percent and below the backtested baseline.
Total STRONG editorial-sized: 9-9. Yesterday added zero. The bot won both its STRONG-tier UNDERs we PASSED. The editorial-sized totals tier has not added a sized bet since the May 28 LVA-DAL loss.
Combined STRONG editorial-sized: 21-22 (49 percent). Below breakeven on the season's editorial line.
Per the grader on all bot STRONG cards: spread 16-20 (44 percent), total 17-17 (50 percent). The bot's published bets are now below breakeven on spreads.
The architecture's standing: 5-0 on non-Dallas applications, 2-3 on Dallas. No architecture firings tonight. The next potential firing is Monday's SEA at LVA HIGH spread, which Maya will read tomorrow morning.
The plan tomorrow.
SEA at LVA is the bot's HIGH pick from Saturday's picks file at 7.9 points of edge. The line is SEA plus 14.5 in the morning capture. Seattle is missing Magbegor (still Out) plus the rotation pieces that have been Out for weeks. The matchup math without Magbegor argues for a heavier line than the bot's morning capture implies. The architecture check requires the agent report Monday morning. If the agent confirms SEA late quarters at HIGH tier and Las Vegas's Q4 closing structure has a documented weakness, the architecture fires and the discipline rule's status depends on which sizing path the user chooses.
For tonight, the discipline holds. Three PASS calls. Zero exposure. The seventh consecutive correct silent-agent PASS extends the pattern that has been the framework's actual edge this season.
Talk tomorrow.
[ End Report ]
Share This