A Bigger Week for Featured Players

Featured Players are the picks most of you check first — so this week we gave them a home of their own, and we made the model behind them smarter.

A dedicated Featured Players page

The picks used to live alongside everything else. Now there's a single page built around them, with a few things we're genuinely excited about:

• ★
  The Daily Double — our two strongest plays of the day, combined into one parlay, with its own win/loss and units tracker. Each leg can be any price, but we constrain the combined ticket to a confident-but-worth-it range so it never gets too chalky or too much of a longshot. Tap it for a full technical breakdown of why those two cleared the board.
• ★
  Two locked trackers, all day — an "Early Bird" read locked in the morning and a sharper "Final" read locked closer to game time. Both stay put the whole day with inline status, so you can watch a pick go from upcoming to in-play to hit or miss instead of having it quietly disappear.
• ★
  Postponements handled honestly — if a game gets called off after its picks lock, those picks now stay visible marked "Void" (and excluded from grading) instead of vanishing from the board.

A revamp to how picks are chosen

Before a pick ever reaches you, our model projects every bettable stat for every hitter and starting pitcher in a game, ranks them, and only lets the one to four strongest plays per game through. That keeps the board focused on real edges instead of filling space. This week we sharpened the projections feeding that gate:

• 1.
  Opponent & ballpark context — a hitter's projection is now tempered by the quality of the opposing starter, a pitcher's strikeout and earned-run lines flex with the lineup he's actually facing, and park factors nudge power and run-scoring stats. The matchup finally shapes the number, not just the player.
• 2.
  Advanced "true talent" signals — metrics like wRC+, xwOBA and FIP layer on top of recent form to separate real skill from luck. A hitter squaring the ball up but getting unlucky gets a small nudge up; surface stats hiding shaky peripherals get a nudge down.
• 3.
  A new pitcher angle: Earned Runs Over — we used to only flag pitchers projected to limit damage. Now, in a genuinely tough spot, we'll flag a starter likely to give up runs. Because that's the noisiest pitcher stat, we hold it to a stricter standard than the under: it has to be backed by the arm's underlying skills or a clearly tough offense — not just one ugly recent start — and clear a higher confidence bar to even appear.

Calibration carries into the +EV engine

The Featured model and the +EV engine share the same projection core, so these upgrades land in both. The most important one for +EV is calibration:

• Regression to the mean — a hot or cold 15-game stretch is high-variance, and pricing straight off it makes the model overconfident (the classic "predicts 65%, actually hits closer to coin-flip" leak). We now shrink a player's recent rate back toward their established baseline before pricing an edge, so streak noise stops manufacturing fake value.
• Cleaner experiments — we deliberately keep the new opponent-context layer out of the +EV scan for now, so its live calibration test isn't confounded by two changes at once. The +EV board stays an honest read on one thing at a time.

What's next

• NBA player props — bringing the same gate + calibration to basketball
• Locked-once-shown picks — guaranteeing a featured pick never rotates out mid-day on busy slates
• More pitcher and matchup signals — extending the opponent layer with Statcast quality-of-contact

— The Neural Network Team