How to read GameTime Picks
A 2-minute guide for everyone — no betting background needed. Everything here is educational and paper-only: there are no real wagers, just hypothetical paper tracked honestly.
Start here
The three probabilities
Wherever we show a chance for the same outcome, you may see it three ways. They are not three opinions of equal weight — the third one is the benchmark the other two are measured against.
What the simulation produced, unmodified. Kept as evidence and never overwritten — and historically about nine points more confident than the settled results justified.
The raw number corrected against what actually happened on earlier slates. This makes our stated confidence more accurate; it does not add predictive information.
Derived from both sides of the posted price with the bookmaker's margin removed. It is their number, not ours, and on the settled record it is the better estimate.
The difference between our number and theirs is a disagreement, not an advantage. On the settled record, the largest disagreements have settled worse than the small ones, so nothing on this site is ranked or recommended by the size of that difference. Where a market's own measured record sits below break-even, we switch its predictions off and keep the history visible.
The numbers on a card
Shown American-style: −150 means risk 150 (paper) to win 100; +130 means risk 100 to win 130. Combine legs and the odds multiply.
A rate only counts results that have been decided. Pending games are not counted yet, and pushes are listed separately rather than folded in.
Every rate on the site is shown with how many decided results it is over and how wide its uncertainty is. A rate over a small sample is mostly noise.
A slate we refused to settle after an integrity check reads “withheld”. A date where no slate was ever built reads “not produced”. Neither carries a rate, and neither is hidden.
Risk tiers
The three tools
Every market the model has a read on — shown with model %, market %, and the gap between them. A “projection view” is information; not every view is suggested as a card.
The model's actual paper picks for the day, bundled into cards by risk tier. Enter any stake to see the projected paper return. Browse them all in the Parlay Center.
Make your own paper card from eligible legs across sports in the Parlay Center — add legs, set a stake, and see the combined odds + payout live.
Bank Builder
A transparent paper ladder: one disciplined pick per rung, compounding a starting bankroll toward a target. It only advances when a genuinely eligible low-risk card clears strict gates — when nothing qualifies, it waits rather than forcing a pick. Track it on Bank Builder. Paper-only, for education.
By sport
The only sport with a live model. Player-prop simulations — pitcher strikeouts, batter hits, total bases — run against the posted line, plus game-level market context.
History only. The settled record from earlier seasons stays readable, but nothing new is being modelled or published for NBA.
Market-implied only. Win probabilities are read from the posted price; there is no fight model behind them, and no method / distance / round markets.
Closed. The 2026 tournament ran 90-minute regulation only (a Draw was a real third outcome). Kept as an archive of what was published at the time.
Why some picks aren't suggested
This market's own settled record sits below break-even, so we make no prediction in it. The history stays visible.
A player prop needs the confirmed starting lineup before it can be shown as a card leg.
The odds provider doesn't offer this market for this event, so there is no price to read against.
Too few decided results to say anything yet. It is reported and never acted on.
How a market earns its place
Each market carries its own settled record. Until it has enough decided results to say anything, it is reported and nothing more — its numbers are never used to lead a page.
A market clearing half its calls proves nothing. It is scored against the de-vigged market price on the identical settled rows, and so far none of ours scores better.
When a market's hit-rate interval sits entirely below break-even across a large sample, we stop making predictions in it. The history stays on the site and is never placed in a ranked list.
Results come from official box scores. If the event mapping fails an integrity check, the whole slate is withheld rather than partially graded.
Glossary
Our model's estimate that an outcome happens, 0–100%.
The same chance baked into the sportsbook price, margin removed.
Model probability minus market probability — a difference, not a proven advantage. Positive = the model reads it higher than the price.
−150 = risk 150 to win 100; +130 = risk 100 to win 130.
Several picks combined into one card — all must hit; the odds multiply.
A single pick inside a parlay card.
A player prop shown before the starting lineup is confirmed — labeled until lineups post.
A pick the model rates but the sportsbook doesn't price — shown without a paper payout (e.g. some UFC cards).
A projection the model has, but that hasn't cleared our bar to be a suggested card yet.
GameTime Picks is an educational analytics project. Nothing here is betting advice or a recommendation to wager. Every “stake” and “payout” is hypothetical paper, tracked honestly — wins and losses both. For every term (model %, market %, model gap, confidence, no-play, pending) see the Market Guide; for the full model write-up see Methodology; for our stance see Responsible use.