30. The margin Chicago beat Portland by Friday night at the United Center. Final CHI 124, POR 94. The bot's POR HIGH pick at plus 5.30 lost the cover by 24.7 points. That is the second straight POR-at-CHI matchup the bot has elevated to HIGH and lost by 20-plus points. The first was Tuesday's 23-point loss (CHI won 101-78). Cumulative editorial damage from the POR-at-CHI matchup over the last seven days: minus 1.0 units at the half-stake default.
The bot's model has been wrong about Portland against Chicago at home. Two consecutive 20-plus-point blowouts the wrong way is a meaningful calibration signal. The way the model rates Portland (the team's per-game data through June suggests they are an average to below-average road team) is not aligned with how Portland has actually performed on the road against quality opposition at home. The model says POR plus 5 is the right read; the market says POR plus 5 is too generous; the games have said the market is closer to right by 20-plus points twice in a row.
Going forward, the framework adds a "POR-on-the-road" filter. When the bot publishes a HIGH-confidence call with Portland as the road team picked, we size at quarter stake instead of half stake until the model recalibrates. The discipline costs ~50 percent of the expected value on the Portland road bets but caps the downside on the matchups that have been the framework's worst observations of the last month.
Saturday's pick was the bot picking Phoenix on the road at Toronto.
PHO at TOR HIGH at minus 8.85 (the actual line by game time; the card was generated when the line was minus 3, which was wide of the closing line). Phoenix won 89-80, covered the closing line by 0.15. The card's spread calculation against the minus 3 opening line said cover by 6, which is the conservative editorial read (we mark winners against the card line, not the closing line, since the card is what we wrote against).
Either way the bot was right. The Tempo without Sykes against a quality defense is a meaningfully worse team than the model implies, and the bot's PHO pick captured that asymmetry. At half stake the bet returns 0.45 units of profit.
Net weekend editorial damage.
Friday: minus 0.5 units (POR HIGH loss). Saturday: plus 0.45 units (PHO HIGH win). Net: minus 0.05 units. Essentially flat over the two-day slate.
Cumulative per the grader through Saturday.
Spread STRONG (all bot-labeled STRONG LEAN cards): 32-30 (51.6 percent, minus 1.47 percent ROI). Total STRONG: 31-31 (50 percent, minus 4.55 percent ROI). The system is still effectively flat after 130-plus observations.
Tonight is four games with no HIGH-confidence picks.
The picks file shows zero HIGH-confidence calls (n_picks equals 0) on a four-game slate. The cards have STRONG LEAN labels on POR at WAS (POR plus 6.5, edge plus 11.36, the largest spread edge of the night) plus STRONG UNDER signals on LVA at CHI and MIN at DAL. Plus a MODEST tier on NYL at GSV. The bot did not elevate any of them to HIGH per its calibration thresholds.
The framework rule from Issue 42 is "only bet HIGH picks at half stake by default; non-HIGH cards stay PASS unless conviction architecture clearly fires." Tonight no conviction architecture fires (no road-team-favored-by-model with agent-confirmed late quarters against a structurally weak home team). Per the rule, PASS all four.
The case for taking the POR plus 6.5 at WAS card despite no HIGH label: edge plus 11.36 is among the largest of the season, the new POR-on-the-road filter says quarter stake on these. The case against: Washington at home has been strong; the bot specifically did not elevate this to HIGH despite the edge size, which means the model has a calibration concern about Portland or about WAS-at-home that the public edge calculation does not capture.
I lean discipline. Quarter stake on POR plus 6.5 at WAS only if the framework's no-HIGH-no-bet default is overridden, which the discipline says it should not be for cards the bot specifically did not elevate. PASS the spread.
The STRONG UNDER signals: LVA at CHI UNDER 178.6 edge minus 15.67 is the biggest under edge tonight. The loosened totals rule says half stake when either pace or injury condition met. Both teams' injury reports do not jump out as cleanly thinned (LVA is missing two rotation pieces but they have been managing it; CHI is mostly healthy). The pace projection lands roughly in line with both teams' season averages. Neither condition cleanly met. PASS.
Tonight's recommended action.
PASS: All four games.
Zero exposure tonight. The bot returns with whatever Tuesday's slate produces.
The Tempo are 11-8 after the Phoenix loss. Their next game is Tuesday at home against Indiana. The four-game test stretch Jordan wrote about Friday has produced one loss to a top-tier defense (Phoenix). The remaining three games are home against Indiana, at Phoenix (rematch), and a TBD game later in the week.
Three framework adjustments shipped this week worth tracking.
One. Model-versus-agent conflict rule loosened (Issue 42). PASS only at 3-plus HIGH agent signals on the opposite side. Not yet tested at the loosened threshold.
Two. Tempo bias correction (Issue 42). Treat Tempo bets at the same threshold as any other team's. Two applications since the rule shipped: TOR plus 13.5 PASS we should have taken (Tempo covered), and PHO minus 8.85 HIGH we should have taken half stake going against the Tempo (we did, won). Net positive on the bias correction so far.
Three. POR-on-the-road quarter-stake filter (this issue). New, tested zero times.
Three rule changes in seven days is more framework adjustment than I am comfortable with from a sample-size standpoint. The July ledger needs to evaluate whether these adjustments produce materially better results without giving back the bankroll safety the earlier rules provided. If the adjustments produce a 55-plus percent run on STRONG-tier bets through July, they stay. If they produce 50-percent variance with bigger swings, the rules get tightened back up.
No editorial action tonight. The bankroll sits. The Tempo Report from Jordan covers Saturday's loss.
Talk tomorrow morning if there is Monday slate news, otherwise Tuesday for the Indiana game.
[ End Report ]
Share This