6.5. The Tempo's cover margin Sunday at Atlanta. Final ATL 94, TOR 87. Toronto lost the game by 7 but covered the 13.5-point spread by 6.5. The bot's only HIGH pick of the night. We passed per the discipline framework — Sykes was Out, the agent had two HIGH Atlanta quarter signals, the model-versus-market gap was 17.25 points, and the calibration with the named star out felt wrong.
The bet would have won by 6.5. At half stake by the default rule, that is 0.45 units of profit we did not collect.
The pattern is now real. Counting back to early May, the framework discipline has passed on at least five or six published HIGH-confidence picks that ended up covering. The cumulative cost is approximately 3.5 to 4.5 units of missed profit. The bot's published HIGH picks are running roughly 56 percent per the grader's full STRONG ledger (which includes a couple of MODEST escalations the bot promoted to HIGH later in the season). At 56 percent on minus 110 spreads, the breakeven plus expected-value math says we should be sizing those bets, not passing them.
The honest update.
The framework needs to be more aggressive on bot HIGH picks. The half-stake default rule was the right adjustment in mid-May. The PASS criteria we layered on top (Vegas Pattern when severe, model-versus-agent conflict when 3-plus HIGH agent signals, named star absence) have been too restrictive in practice. The framework is producing 50 percent on the STRONG-tier bets we do place because we are only placing the ones with the cleanest signals, while the rougher-signal bets we pass have been hitting at roughly bot-publication-tier rates.
Two corrections going forward.
First. The model-versus-agent conflict rule fires at PASS only when the agent has 3-plus HIGH signals on the opposite side. Two HIGH signals plus one HIGH signal on the picked side is no longer a PASS — it is half stake by default. Sunday's TOR matchup had two ATL HIGH (Q2, Q4) and one TOR HIGH (Q1). That is two-versus-one. The rule says PASS at 3-plus, half stake otherwise. The named-star-out additional filter was a discretionary brake that has now been wrong on two consecutive applications. Loosen it back.
Second. The Tempo bias correction. We have been passing Tempo bets at a higher rate than other teams' bets because of the editorial-team conflict (the Tempo are the team Jordan covers most heavily, the team most of our readers care about). The bets we have passed because the Tempo are involved have been costing us money. Going forward, treat Tempo bets at the same threshold as any other team's bets per the framework filters.
Cumulative through games Tuesday morning.
Spread STRONG (per grader, all bot-labeled STRONG LEAN cards): 27-26 (51 percent, minus 1.05 percent ROI). Up one game from yesterday after the Sunday TOR cover the bot was right on. Total STRONG: 25-24 (51 percent, minus 2.60 percent ROI). The bot's HIGH-confidence-published-picks subset (the n_picks list, not all STRONG LEAN labels): roughly 56 percent over the season per the grader's tracking. Above breakeven.
Tonight is 4 games. Two HIGH picks.
Phoenix at Indiana. Picks file: PHO HIGH plus 6.1, edge plus 6.10. Model has PHO favored by approximately 1 point at neutral. The market has IND favored by 6.1 at home. The model thinks PHO is meaningfully better than the line implies.
Agent check. The agent has IND Q2 plus 3.20 HIGH (Indiana opens the second quarter strong at home) and IND Q3 plus 2.00 MEDIUM (Indiana keeps the lead). One HIGH agent signal favoring the home team, no HIGH signals favoring the picked road team. That is not the 3-plus HIGH agent threshold for PASS. The framework default per the loosened rule above is half stake.
Half stake on PHO plus 6.1 at Indiana.
Portland at Chicago. Picks file: POR HIGH plus 5.65, edge plus 5.65. Model has POR favored by approximately 1 point at neutral. The market has CHI favored by 5.65 at home. Same shape as the PHO call.
Agent check. The agent has POR Q1 plus 2.50 HIGH (Portland opens strong) and CHI Q2 plus 2.50 HIGH (Chicago answers in the second). One HIGH each side. The SGP correlation reads "Close game, POR closes." The agent profile is tied. The framework default per the loosened rule is half stake.
Half stake on POR plus 5.65 at Chicago.
The other two games.
Golden State at Atlanta. Picks file has no HIGH call. Card has PHO label as PASS or MODEST per the bot's threshold. Agent has 3 HIGH signals (GSV Q1, ATL Q3, ATL Q4) with the ATL signals being the larger of the three. Architecture lean is Atlanta at home but no published HIGH bet. PASS per the framework — only bet HIGH picks at half stake by default; non-HIGH cards stay PASS unless conviction architecture fires.
Washington at Minnesota. Picks file has no HIGH call. Card has PASS spread (model edge minus 0.8, agrees with market). Card has MODEST UNDER 167.5, tier downgrade per discipline.
But the agent's read here is interesting. All four quarters have HIGH signals favoring Minnesota (Q1 plus 3.60, Q2 plus 5.60, Q3 plus 3.10, Q4 plus 2.50). That is a complete agent dominance pattern for the home team — the kind of architecture-fires signal we usually wait for. The reason the bot did not elevate this to a HIGH pick is the model's spread edge is essentially zero (model has MIN minus 5.0, market has minus 5.8, edge minus 0.8). The agent says blow them out; the model says it is roughly priced.
When the agent and the model agree on direction but the model says it is priced, we do not bet. The framework requires both the directional read and the edge size to clear the threshold. Tonight MIN at WAS is the right side per the agent but not enough edge per the model. PASS.
Tonight's recommended action.
HALF STAKE: PHO plus 6.1 at Indiana. HIGH-confidence published pick. Default rule.
HALF STAKE: POR plus 5.65 at Chicago. HIGH-confidence published pick. Default rule.
PASS: GSV at ATL (no HIGH pick, agent leans home).
PASS: WAS at MIN spread (no edge per model).
PASS: WAS at MIN total (MODEST tier).
Total exposure tonight is roughly 1 unit (2 half stakes). Expected value if the half-stake default holds at its sample-implied 55-56 percent rate is plus 0.13 units per pick, or about plus 0.25 units for the slate. Variance is high on a two-pick night.
The framework is in a recalibration window. The discipline is producing 50 percent on what it bets and missing 56 percent on what it passes. The loosened model-vs-agent rule from tonight is the first step. The Tempo bias correction is the second. The June ledger is essentially flat; July needs to be the test of whether the loosened framework produces materially better results without giving back the bankroll safety the original rules provided.
Tempo plays tomorrow night at home against Las Vegas — Jordan will have the matchup preview in the morning. No Tempo game tonight, so no Tempo Report.
Talk tomorrow morning.
[ End Report ]
Share This