How GetCloseline.io works: data, model and grading

GetCloseline.io is an AI sports betting model. It prices games across NBA, NHL, MLB, NFL, NCAAF, NCAAB, EPL and MLS, flags only the bets where its estimate of the true probability is better than the sportsbook price implies, and sends a pick only after a separate AI review step approves it. Every pick is timestamped when it is sent, graded automatically from official final scores, and kept on a public record, losses included.

As of October 30, 2026, the public record is 409-301-17 (wins-losses-pushes), a 57.6% win rate over 710 graded picks, counting every pick sent since June 6, 2026.

What it covers

The model runs on NBA, NHL, MLB, NFL, NCAAF, NCAAB, EPL and MLS. For each league it prices the moneyline, the spread and the total (over/under), and it prices player props in MLB, NBA, NHL, NFL and college sports where books post them. Coverage follows the calendar: college football runs August to January, college basketball November to April, and a league produces nothing while it is out of season.

Not every market that is priced is published. Each league-and-market pair is run in shadow first, which means it predicts and is graded but nothing is sent, and it is only published to members once its shadow results have earned it. A pair that stops performing is moved back to shadow automatically.

As of October 30, 2026, the leagues with graded picks on the public record are EPL, MLB, MLS, NBA, NCAAF, NFL and NHL.

How a pick is produced

• Data. Odds from several sportsbooks, schedules, injury reports and team results are collected every morning before the day's games.
• A fair price. For each game the model estimates the probability of each outcome from team strength, matchup, recent form and situation, independent of the posted odds.
• The market price, without the margin. The sportsbook's built-in margin (the vig) is removed so the model's number and the market's are compared like for like.
• An edge, or nothing. A candidate exists only where the model's estimate beats the price by enough to matter. Most games produce no pick, which is the intended result.
• Filters per market. Each league-and-market pair has its own limits on price and confidence, learned from that pair's own graded history rather than set once for everything.

The Pick Critic

Before anything is sent, each candidate goes to the Pick Critic, an AI review step built on a large language model. It reads the pick alongside its context, such as injuries, the matchup and the price, and either approves it or vetoes it with a reason. Vetoed picks are still recorded and graded internally, so the critic itself is measured: if the picks it rejects start winning more often than the ones it approves, its threshold is loosened automatically.

How the system corrects itself

Every night, after games settle, automated checks compare each league-and-market pair with its own history. A pair that is getting worse is diagnosed, adjusted or pulled back to shadow; one that has recovered can earn its way back. If the overall win rate collapses over a week, the morning broadcast pauses itself until it recovers. These checks run on a schedule with no person in the loop.

How results are graded

When a game is final, each pick is marked WIN, LOSS or PUSH from the official final score. Win rate is wins divided by wins plus losses; pushes are shown but are not counted in the rate. Nobody grades picks by hand.

Every pick is timestamped when it is sent

Each pick is stored when it is sent. The public pick log lists every settled pick with the time it was sent and the scheduled start of the game, and the same data is available as JSON at https://www.getcloseline.io/api/public/record.json. Of the 727 picks on the record as of October 30, 2026, 725 were sent before the scheduled start; the 2 sent at or after the start are listed separately rather than hidden.

What counts toward the record

• Only picks that were actually sent to members. A pick the model made but never sent is not a win or a loss, because nobody received it.
• Only picks the Pick Critic approved.
• Only the current model version for each market, so an old version cannot pad the numbers.
• Every loss. Nothing is deleted once it is graded.

What we do not publish, and why

We publish win rate and record only, not a stake-based return figure. Members size and choose bets differently, some take every pick and some take a few, so a single running return would describe a bettor who does not exist. Win rate and record are the same for everyone who received the picks.

That has a cost worth stating plainly: a win rate on its own does not prove an edge. Picks on favorites win more often and pay less, so a high win rate can still lose money at short odds. Judge the record alongside the prices you were able to get.

Where to check it

• Track record: win rate and record by league and bet type.
• Pick log: every settled pick with its send time and the game's start time.
• FAQ: price, the free version and how the record is kept.
• Today's free picks: a free selection each morning.

GetCloseline.io Premium is $9.99/mo after a free trial (first month free).

Analysis and entertainment only, not betting or financial advice. No model wins every bet and past results do not predict future ones. 21+. If you or someone you know has a gambling problem, call 1-800-GAMBLER.