MLBQUANT

Every call logged before first pitch. Graded after.

A student-built MLB model that compares its win probabilities against the betting market, bets on paper under an eight-gate policy, and publishes the graded ledger — wins, losses, and everything it refused to bet.

Record
0-1
P&L
-1.0u
ROI
-100.0%
Pending
14
Date
Pick
Odds / Model
Grade
Units
2026-07-17
BAL ML (Baltimore Orioles @ Houston Astros)
+101 / 56%
·
pending
2026-07-17
STL ML (St. Louis Cardinals @ Arizona Diamondbacks)
+104 / 60%
·
pending
2026-07-17
CIN ML (Cincinnati Reds @ Colorado Rockies)
+101 / 58%
·
pending
2026-07-17
MIN ML (Minnesota Twins @ Chicago Cubs)
+130 / 55%
·
pending
2026-07-17
NYY ML (Los Angeles Dodgers @ New York Yankees)
+100 / 58%
·
pending
2026-07-17
TB ML (Tampa Bay Rays @ Boston Red Sox)
+104 / 56%
·
pending

Eight gates between a flag and a bet

The model disagreeing with the market is not enough. Every prospective bet passes this checklist or is published as suppressed, with the reason — a system that hides its rejections can't be audited.

  1. Minimum edge 3%Below that, model-vs-market disagreement is noise.
  2. Expected value at the real priceThe edge must survive the vig — profit is computed at the posted odds, not the fair line.
  3. Probability band 30–70%The side doesn't need to be favored, but it must sit where the model's calibration is proven.
  4. Both starting pitchers postedA flag built on placeholder pitching isn't the model's real opinion.
  5. Odds between −250 and +250Beyond that, “edges” are usually stale lines or payout traps.
  6. Edge sanity cap 15%Too good to be true means a data bug, not market-beating insight.
  7. Game not startedA bet logged after first pitch is hindsight, and hindsight poisons a track record.
  8. Model must beat its baselineThe totals model failed this check — so the system refuses its bets, automatically.