May 20 — 0-1, moneyline loss cost half a unit

Yesterday

0-1 · -0.50u · ROI -100.0% · CLV +2.33%

Rough start to the slate with Miami falling 1-9 to Atlanta despite the model finding 10.48% expected value at 41% win probability versus the market's 36%. The thesis completely backfired as Chris Sale's 69.20 rating dominated Janson Junk's 38.00 rating even more decisively than projected, while Atlanta's four position player injuries proved irrelevant. The 2.33% closing line value suggests the read was sharp, but Sale's excellence and Miami's offensive collapse made it academic. Sometimes the better pitcher just takes over.

What missed

Worst result. MIA ML (ATL @ MIA) at +170, -0.50u — final 1-9, CLV +2.3%. Graded unlucky (positive CLV, lost).

The model catastrophically misjudged Janson Junk's performance, predicting 4.13 xFIP but he allowed 8 ER in 5 IP with 2 HR. Chris Sale delivered as expected with 1 ER over 7 IP and 8 K matching his 30.9% K rate projection. Miami's bullpen score of 65.0 proved accurate as they held the lead. The model's 7.74 total runs badly undershot the 10 actual runs. With 2.33% CLV, this was an unlucky loss on a fundamentally sound bet. Retrain priority: starting pitcher volatility modeling.

Under the hood

Yesterday's development focused on improving betting line value tracking and website reliability. The team enhanced how the system calculates profit margins on run line bets, fixing gaps where certain picks weren't being properly analyzed. They also upgraded the website's security features and refreshed the visual branding with a new logo set. Additional work included refining automated social media posting and strengthening the system's ability to audit pick quality after games conclude.