MLB Analytics Blog · Sep 10, 2026 · 8 min read
How the Sportslyx MLB board is built each day: which games qualify, why it reads player props under the line, how a read is chosen and frozen, and how it is graded.
By Sportslyx · Published
Search for AI MLB predictions today and most of what comes back is a list of picks with a percentage and no explanation of where either came from. This article is the explanation for the Sportslyx MLB board: which games are eligible, which market is read, how a read is chosen, how it is frozen, how it is graded from the box score and how the result reaches the public record. Everything below describes the code that runs the board. Nothing here is a claim about how well it does — that lives on the record, not in an article — and today’s reads are not in the article either: they are on the board for subscribers, and yesterday’s are on the Results page for everyone.
For every other sport the company board reads the moneyline favorite. Baseball is the exception. The MLB board reads player props exclusively, and a game with no eligible prop line simply has no read — there is no fallback to the moneyline. The reason is what a baseball moneyline looks like: the favorite typically sits at a fair probability in the low-to-high fifties, which is a coin flip at a price that already carries the bookmaker’s margin. Switching the market does not create an edge — a prop pair is de-vigged the same way and carries the same margin — but it gives the board a read whose probability and price are on a scale the record can be built from.
The prop lines come from a public core feed, in Over/Under pairs for a player, a statistic and a line. Nine statistics are read, and they are exactly the nine the box score can settle: for pitchers, strikeouts, outs recorded, walks allowed, earned runs allowed and hits allowed; for batters, hits, walks, runs and RBIs. Total bases, singles, doubles and one-sided milestone markets are left out on purpose — the box score does not separate doubles from triples, and a milestone has no other side — because a read that cannot be graded is worse than no read.
A scheduled job walks the current day’s schedule and the following day’s, once every couple of hours through the day, and tests each game against the same structural rules before any price is looked at:
Prop menus are published on game day, which is why the following day’s games pass the first four tests and fail the last: they are visited, found to have no lines, and picked up on the day itself. A game the job has already placed a read on is left alone on later passes.
The band is there for two different reasons. Below 60% a prop is no better than the moneyline favorite it replaced and harder to grade. Above 84%, and beyond −500, one miss costs many wins, and the price no longer pays for the risk that the line and the box-score row do not match. No language model takes part in any of this. The written breakdown for the game is produced separately and cached; the read is chosen by arithmetic on the lines, and the player’s name is looked up only for the line that was chosen, because a game can carry a hundred candidates.
This is the one baseball-specific rule on the board, and it comes from measuring the board’s own settled reads rather than from theory. Every statistic the board reads is a counting statistic — hits, walks, runs, RBIs, strikeouts, outs — and every one of them depends on how long the player is on the field. A batter lifted for a defensive replacement, rested for the second game of a doubleheader or pulled in a September rotation gets fewer plate appearances; a starter on a pitch limit records fewer outs and fewer strikeouts. Playing-time risk pushes every counting statistic down. The Under side is protected by that risk; the Over side is exposed to it.
The grading rule sharpens the asymmetry. A player who never appears is void, but a player who appears once is graded on that appearance, so “Over 0.5 hits” for a batter who gets one at-bat is nearly always a loss while “Under 0.5 hits” for the same batter is nearly always a win. Measured across the board’s settled reads, the Over side had been losing markedly more than the Under side at similar prices, and the rule removes that segment.
When the job places a read it snapshots the line, the price and the fair probability onto the graded row, and a later pass does not re-snapshot a line that has drifted — the record has to grade exactly what the card showed. The card is published to the board with its label and its probability, which is the prop’s de-vigged figure, not a model’s opinion. Nothing is edited afterwards, and a game that has passed the cutoff is never added retroactively.
A settlement job runs every hour. It waits for the game to be final and then, because the read is a prop, opens the official box score rather than the score line. The player is found by the same athlete identifier the prop line carried, so “Luis L. Ortiz” and “Luis Ortiz” cannot be confused; name matching survives only as a fallback for reads placed before the feed carried identifiers. The job reads the column for the statistic — outs are derived from the innings-pitched figure, so 5.2 innings is seventeen outs — and compares it with the line. Every line is half-numbered, so the result is a win or a loss and a prop can never push.
If the player has no row in the box score, the grader returns nothing rather than zero. That distinction is the most important line in the grader: a zero would make every Under win automatically for a scratched player. The read stays pending instead and is voided if it has not resolved within three days.
Each settled read is booked at a flat $100: a win at −250 books +$40, a loss −$100, and the price used is the one frozen when the read was placed. Hit rate is wins divided by wins plus losses; ROI is the net result divided by the total staked; the period is one calendar month in UTC. The same functions produce the figures a subscriber sees and the figures on the public pages.
Subscribers see today’s reads on the board as they are placed. The public Results ledger shows every settled day in full — the player and line, the price, the probability, the outcome and the result at a flat stake — while today and the following days appear as counts only. The Pick Tracker carries the monthly record by sport, and the MLB predictions board shows today’s games with the market-derived probability of the moneyline favorite — a different number from the prop read, which is why the two are never printed as one — above the same monthly MLB record. Individual games, with lines and probable starters, are indexed under MLB games.
Past results do not predict future results, and nothing on the board is advice. It is a method whose every step is written down, run by a job, graded from official data and published — so it can be checked rather than believed. For the rules that apply across all sports, see the methodology; for the numbers, see the record.
No. MLB is the one sport where the company board reads player props only. If a game has no prop line that passes the filters, there is no read on that game — the board does not fall back to the moneyline favorite.
Because the read is a player prop, such as a starter under a strikeout line or a batter under a hits line. It is graded from that player’s row in the official box score after the final out, not from which team won.
A player who never appears has no box-score row, and the grader returns nothing rather than zero. The read stays pending and is voided if it never resolves within three days. A player who appeared even once is graded on what he did.
Prop menus are published on game day, so the job that places reads finds nothing for tomorrow’s games and returns to them on the day. It runs through the day and places a read once eligible lines exist and before the cutoff, fifteen minutes after the scheduled first pitch.
Take this guide to the MLB board: the 7-day Pro trial opens the AI breakdowns, the market-derived probabilities, the daily company board, Combo Builder, Studio and the pick tracker.
7-day trial · $0 today · card required · cancel anytime before the trial ends
Sportslyx produces statistical estimates and analytical breakdowns for information and entertainment. Nothing on this page is betting advice, and no outcome is certain. 18+ only. Please play responsibly — see our Responsible Play page.
