Learn the words in order.

Always visible

How we compile the numbers

This is the cross-check behind every win chance on AI Picks, Games, and Parlay. Read it once. It does not change unless the board does.

1. 01 · Live schedule, not a made-up slate

   Kickoff, home, away, and league come from ESPN's live scoreboard for NFL, MLB, college football, NBA, and NHL. If a game is not on that board, we do not invent it. Confirm the fill at Hard Rock Bet — public odds are delayed.

2. 02 · Strip the sportsbook's cut — that is the prior

   A −110 / −110 market is not 50/50. We convert both prices to implied chances, remove the extra so they add to 100, and treat that no-vig close as the most precise look. Liquid sportsbooks are extremely hard to beat with public data. We do not pretend otherwise.

3. 03 · Prediction markets are a second crowd

   Kalshi event contracts and Polymarket moneylines are real money from a different venue. We take the mid-price, then down-weight thin books (wide bid/ask, tiny volume) and up-weight liquid ones. They are research. They are not a Hard Rock ticket and they never steal today's pick.

4. 04 · Convert every other number into a win chance

   The point spread becomes P(win) with a sport-specific scoring curve (NFL σ ≈ 13.5 points, MLB ≈ 3 runs). The opening line is a second sportsbook print. ESPN's matchup model is an independent rating. We do not double-count a spread that already agrees with the moneyline.

5. 05 · On-field models: log5, Pythagorean, pitchers, margins

   Season or home/road records go through Bill James log5 (with home-field only when the split did not already include it). Points scored and allowed become a Pythagorean win rate that catches lucky records. MLB starters are blended at ~61% of innings with the bullpen, using ERA and WHIP. Last-10 score margins are recency-weighted — a one-run luck win is not a three-score beatdown. That is analysis of the live log, not a generated card.

6. 06 · Context is light on purpose: rest, injuries, series, park, weather

   NBA/NHL back-to-backs, NFL byes, Out/IL counts (questionable is discounted), this series' head-to-head, and extreme MLB parks get small votes. Weather moves totals more than winners; we do not flip a moneyline on wind. News is filtered to this game's two clubs and never enters the number.

7. 07 · Pool in log-odds. Shrink when they fight.

   Each look has a precision (how much we trust it). We average in log-odds, weighted by precision. If the looks disagree, non-market layers are scaled down — a hierarchical shrink toward the sportsbook. High totals add chaos, so we pull slightly toward 50/50. The result is capped so we never print a fake lock. A parlay multiplies; it does not average.

8. 08 · The call has to pay, and it has to be today

   We rank tickets by researched chance and how much a $1 bet actually returns. A −400 favorite that wins 80% of the time still pays almost nothing, so it does not jump the line. AI Picks only uses games tipping today — popular, player, period, and parlays. Photograph the Hard Rock Bet Florida number before you confirm so the live price replaces the delayed print.

Research, not a promise. Delayed public odds. Confirm every live number at Hard Rock Bet. 21+ · 1-800-GAMBLER.