Two things live here: how the model is validated, with the receipts, and how the Research desk turns news into coverage. One rule governs both: we publish the evidence that it works, never the machinery that makes it work.
Big Mike’s engine prices every line against its own fair number: de-vigged across a dozen sportsbooks, then adjusted by a custom, proprietary model more than three years in the making, tracking weighted variables down to the position and personal-background level. A pick exists only when the posted price beats that number. You don’t have to take the machinery on faith, though. You can watch it get graded. Everything below is out-of-sample: the model makes its call using only what it could have known at the time, then reality grades it.
The reason a model has to exist at all: raw single-season stats are mostly noise. Take batter-vs-pitcher-handedness splits: a hitter’s “big platoon advantage” from last year swings about 4x wider than what actually repeats the next season. Bet the raw number and you’re betting the noise. The engine discounts small samples toward what the evidence supports, and lands on top of reality.
Discounting isn’t enough. The ordering has to mean something. Sort every matchup into five groups by the model’s projected edge, then check what happened a season later: the payoff climbs group by group. When the model says a matchup is worth more, it is worth more.
Graded head-to-head on those 1,150 matchups, the model beat “just use last year’s split” on 63% of calls, and beat “ignore the matchup entirely” on 52%. We publish the 52% on purpose: that margin is thin, it compounds over thousands of priced lines, and a methodology page that only shows flattering numbers isn’t a methodology page.
A backtest says the method worked; the public record proves it keeps working. Every pick Big Mike texts is graded after the game and posted, wins, losses, and closing-line value alike, with nothing deleted. The record is the ongoing version of every claim above.
The inputs, the features, the weights, and the sizing math never publish. That’s the product, and saying so plainly beats pretending otherwise. What you get instead is the only thing a black box owes you: out-of-sample validation, honest baselines, and a public graded record you can audit any morning.
When the fair number and the posted number disagree enough, a play exists. Here’s one settled example day, plays identified by market only. The current slate is the product, and it stays in the texts.
The desk tracks curated feeds across the NFL, MLB, college football, NBA, and WNBA: national insiders, beat coverage, and market-moving news, refreshed every 15 minutes, around the clock. When the same story surfaces across multiple independent feeds within a short window, it gets priority. That cross-source signal, not any single headline, decides what we cover.
Big Mike’s engine de-vigs the market across sportsbooks to compute fair odds, expected value, and Kelly-criterion sizing for every pick he sends. Coverage on this site draws on the same numbers. That layer is the point: the news you can read anywhere; what it does to the line is the part we add.
Every piece is drafted at Big Mike’s Desk from sourced material and the desk’s own numbers, under one hard rule: never invent a stat, a quote, an injury, or a line. Every draft then waits in a review queue where an editor approves, edits, or kills it before anything publishes. Sources are linked at the bottom of every article.
Big Mike’s picks are graded after every game: wins, losses, and closing-line value. The record is public. No deleted picks, no cherry-picked screenshots. If the reads stop earning their keep, you’ll see it in the numbers before we tell you.
Sports betting is entertainment with variance. Please bet responsibly, and only if you’re 21+ where it’s legal. Questions about the process: reach the team.
GET BIG MIKEHis whole board live, your bets graded against the close, and the best number on every play. 61.5% on the moneyline board · 808 picks graded in public.