Strikeo Strikeo

Methodology: how the Strikeo model works

Strikeo is an MLB analytics tool. Every day the model evaluates each scheduled game, compares its own estimate against the betting market, and only publishes a pick when the numbers genuinely justify it. This page explains, in plain language, how that process works and how we verify our results.

What the model looks at

Each game is scored across several groups of signals built from public MLB statistics:

  • Starting pitching. Beyond traditional ERA, the model leans on estimators that better capture true skill — xERA and SIERA — together with strikeout rate (K%) and command indicators (walk rate, quality of contact allowed). Recent form is weighted so a starter's last few outings matter more than a stale season-long average.
  • Offense. Team hitting is measured with wRC+ and OPS, including splits versus left-handed and right-handed pitching, so the matchup against the actual starter on the mound is what counts.
  • Bullpen. Relief quality matters, but so does availability: the model tracks each bullpen's workload over the previous three days to flag tired or depleted relief corps.
  • First inning. Dedicated signals estimate the probability of early runs (NRFI/YRFI), combining each starter's first-inning history with the top of each lineup.

What we publish every day: the game read

For every game on the slate, Strikeo publishes a game read: an analytical breakdown built from explicit, documented rules — a starter whose ERA the estimators don't back, a bullpen that threw 12 innings in three days, a lineup that strikes out a quarter of the time, a park that holds the ball down. Every claim cites its number. The read never comes back empty: when a matchup is even, it says so and explains what was checked. The read is analysis, not a recommendation with a price; its angles carry no expected value and no edge.

Anchored to the market, not against it

Separately, the model prices each market. It does not do so from scratch: it starts from the sportsbook consensus, removes the bookmaker's margin (the vig) to obtain the no-vig implied probability, and lets its own evidence adjust that baseline up or down. The difference between the adjusted probability and the market's is the edge. A value pick is only published when the edge exceeds a per-market minimum (2.2% to 3.5%) and the expected value (EV) at the available price clears its own threshold. If either fails, nothing is published for that market.

The honest part: value is rare

We measured it. Across the full 2026 season (about 30,000 evaluated candidates, settled against official results), the model's probabilities are statistically indistinguishable from the market's — it does not predict outcomes better than the closing price does. That is why, most days, there is no value pick at all, and why we will not lower the thresholds to manufacture one: every configuration we tested that publishes more picks loses money to the vig. When the model passes on a game, the page says so plainly: the price is fair. The game read is what you get every day; a value pick is the rare exception, flagged only when it genuinely appears.

One thing the model does do measurably well is timing: its evidence tends to point the way the line will move between open and close. Where that signal is strong, the read marks an angle as "take it early" or "wait" — an observation about price, never a promise of profit.

Verification and track record

Every published pick is settled automatically against the official MLB result — no manual grading, no silent deletions. The complete history (record, ROI, and units won or lost) is public on the track record page, starting from the cutover to model v2. We also measure CLV (closing line value): whether the prices we published beat the market's closing line. CLV is a quality control on the process itself, independent of short-term win-loss luck.

Limitations

Baseball has enormous variance: a well-founded pick can lose, and a bad one can win. Some signals rely on short samples (a starter's recent outings, early-season team stats) and carry real uncertainty. Strikeo publishes probabilities and reasoning — never certainties. No result is guaranteed, and nothing on this site is betting or financial advice. If you bet, do so legally, responsibly, and only with money you can afford to lose.

New to the terminology? See the glossary for definitions of edge, EV, CLV, and every other term used on the site.