16. The margin Atlanta beat Connecticut by last night. The number the model-agent conflict rule was right by. The bot card had the spread as a STRONG LEAN on Connecticut at the road favorite minus 14.6. The agent had four HIGH-confidence quarter signals all on Atlanta's side with an "ATL dominant game" SGP correlation. The rule said PASS. Atlanta won 91-75.
That is the conflict rule going from 2-2 to 3-2 in a single game. The rule's record now sits at 60 percent across five firings. Two of the three correct PASS calls have been on the kind of qualitative split the rule was designed to catch (agent has hard evidence on the opposite side of a marginal bot pick). Last night's firing was the most extreme version of the split the rule has produced.
Portland at Golden State was the second framework save of the night.
The bot card had POR as a STRONG LEAN at the road favorite minus 8.4 with an edge of 9.51 points. The agent was completely silent — zero quarter edges, zero player props, zero SGP correlations. The conviction architecture did not fire (POR was not favored at neutral; the model had GSV plus 1.11 at neutral). This was a half-stake candidate by the partial-confirmation rule.
We PASSED per the reasoning that the half-stake tier was dragging the season's ledger and a silent agent meant no confirmation of the bot's lone read. Final score Golden State 95, Portland 77. Golden State won by 18. The half-stake would have been 0.5 units lost. The full-stake architectural version (which was never on the table because the architecture did not fire) would have been 1 unit lost. The PASS saved the entire spread of editorial-sized risk on the game.
The total on the same game was MODEST UNDER 162.5. Final 172 went over. The MODEST UNDER would have lost at any size. We PASSED by tier rule (MODEST UNDERS without structural confirmation are PASS by default).
Combined the framework saved at least 1.5 units last night across the two PASS calls on bot STRONG LEAN cards. The discipline has now produced the correct PASS on four consecutive STRONG-tier reads this week (Sunday's LVA at GSV UNDER, Tuesday's CON at ATL conflict, Tuesday's POR at GSV silent-agent, Tuesday's POR at GSV total). The framework's filter set is doing the work it was built to do.
The cumulative ledger through June 2.
Spread STRONG (editorial bets, sized): 12-13 (48 percent). Down a game from Sunday's count. Yesterday added zero to the editorial line because the only STRONG cards (CON and POR) were both PASS calls.
Total STRONG (editorial bets, sized): 9-9 (50 percent). No change.
Combined STRONG: 21-22 (49 percent). Below breakeven on the season for the first time since the early-May run dragged the line below 60 percent.
Per the grader on all bot-labeled STRONG cards including ones we did not size: spread 16-16 (50 percent), total 15-17 (47 percent). The bot's record is converging on the backtested baseline of 57.9 percent on spread STRONG and falling below the baseline on totals. The bot's published bets are doing roughly the same as the editorial-sized version on the season, which is a different conclusion than yesterday's piece (the divergence was based on a shorter sample that has now closed).
The architecture's standing.
Non-Dallas applications: still 5-0. No firings yesterday or today.
Dallas applications: 2-3. Still three straight losses on Dallas applications. The Dallas exclusion rule remains the active framework discipline.
Tonight is PASS again.
Two games on the WNBA slate tonight. The bot's published picks file has only one tonight-eligible call, and it is MEDIUM tier on the PHO at SEA spread (the picks file shows the SEA lean at MEDIUM with edge minus 4.75). MEDIUM tier spreads are not auto-bet by the bot's published rules and not editorial-bet by our framework.
TOR at NYL. Spread label on the card is PASS despite a model edge of 12.97 points on Toronto plus 9.8. The divergence model's label-setter put this at PASS, the picks file's HIGH-tier output is empty, and the agent is silent on quarter projections. Three filters aligned at PASS. We follow.
Total on the same game is STRONG LEAN UNDER 172.6 with edge minus 8.8. The structural-confirmation rule on STRONG UNDERS has the injury condition cleanly met (both teams missing core players: Ionescu Out for New York, Harrison and Fagbenle Out for Toronto), but the pace condition is not cleanly met (New York is upper-third in pace metrics). The agent is silent on the total. Nina has the tactical piece this morning with the specific case for and case against. The verdict is PASS per discipline.
PHO at SEA. Spread MODEST. Total PASS by edge size. PASS both per tier.
Zero exposure tonight. The framework PASS streak goes to five if both PASS calls are correct (or remain neutral, which they will if no editorial bets are sized). The discipline holds for the fifth consecutive STRONG-tier read this week.
Tomorrow's preview.
The slate Thursday already has two cards on the board, and one of them is a HIGH pick from the bot's published picks file. Golden State at Minnesota tomorrow night. The bot's picks file has the GSV spread pick at HIGH confidence with an edge of 6.55 points. The line is GSV minus 2.5 in the market, the model has GSV minus 4.05 at neutral. Plus Napheesa Collier is Out for Minnesota — a star absence that thins the Lynx's offensive ceiling.
Architecture check on the morning data. Road team favored at neutral per the model: GSV is favored by the model at neutral by 4.05 points. The first condition is confirmed. Agent confirmation on late quarters: we will read the agent report when it lands tomorrow morning. Home team Q4 structural weakness: Minnesota's 2025 closing record was middling and the 2026 sample without Collier is going to be tested for the first time tomorrow.
Two of the three architectural conditions look like they will be confirmed. The third requires the agent report we will have tomorrow morning. If the agent confirms GSV in the late quarters at HIGH tier and Minnesota does not show a Q4 closing structure in the data, the architecture fires for the sixth non-Dallas application of the season. Conviction architecture is 5-0 on non-Dallas. The sixth firing would be the season's continuation of the pattern that has been the framework's strongest weapon.
Plus the wrinkle: Golden State is on a back-to-back per the picks file (away_b2b: true). The B2B penalty in the model already discounts the GSV side by minus 2.5 net rating per the betting-model documentation. The 6.55-point edge already accounts for the B2B penalty. The edge is real after the discount.
Tomorrow morning's loop will have the full read. Tonight is PASS. The discipline holds.
[ End Report ]
Share This