May 3 — 0-1, moneyline loss triggers model retrain

Yesterday

0-1 · -0.50u · ROI -100.0% · CLV -4.94%

NIGHTLY RECAP

Baltimore's undervalued 40% implied probability thesis crumbled as the Yankees demolished them 11-3, with Max Fried dominating while Trey Gibson's 4.22 xFIP concerns materialized spectacularly. The anticipated Yankees bullpen weakness (42/100 scoring unit) never factored as their offense exploded early. Baltimore's superior -7 OAA defense couldn't contain New York's offensive barrage. The -4.9% CLV suggests the market correctly priced this mismatch despite our contrarian angle.

Record: 0-1 | -0.5u

What missed

Worst result. BAL ML (BAL @ NYY) at +180, -0.50u — final 11-3, CLV -4.9%. Graded bad bet.

The model's 12% EV pick on Baltimore suffered from systematic bullpen misjudgment. Baltimore's projected 41.9 bullpen score proved catastrophic as relievers allowed 8+ earned runs after Gibson's 4.67 IP/3 ER exit, while New York's 90.0-rated bullpen held firm. The model's 9.70 total runs prediction missed the 14 actual by 4.3 runs, primarily from underestimating Baltimore's relief collapse. Gibson's 2 strikeouts in 4.67 innings fell well short of his 22.5% K-rate projection. The -4.9% CLV suggests this was a genuinely poor bet, not bad luck.

Under the hood

Yesterday's development focused on security and transparency improvements. The team added role-based access controls to limit who can query certain data tables, implemented signature verification for the Discord bot, and established rate limits for AI-generated content. They also built in a 10-day delay before showing AI reasoning to users, ensuring picks can be audited after games conclude. Additional work included better data validation for game start times and system health monitoring anchored to daily slates rather than individual picks.

This week

Last 7 days: 4-3 · +1.30u · ROI +21.7%