MLB prop research in the ET Parlays Lab

ET Parlays rates MLB player props and shows the reasoning behind every rating. It is a research tool, not a sportsbook: it never takes a bet, holds no money, and picks nothing for you. This page is the whole inventory. 11 markets, 16 research surfaces, what actually goes into one read, and the record, including the parts that do not flatter us.

What is underneath one MLB prop

A sportsbook hands you a name, a number and a price. Aaron Judge, 1.5 total bases, -130. Everything that decides whether that is a reasonable bet is missing from the line, and most of it is knowable before first pitch.

Here is what the product assembles for one read, and why each piece moves the number. You are meant to be able to disagree with any of it.

1. The batter’s season. The baseline. How often he reaches, how hard he hits it, how often he strikes out. A short season is regressed toward the league so twenty good games cannot masquerade as a skill.
2. His recent form. The last fifteen days, read on contact quality rather than on results. Batting average over two weeks is mostly luck. Exit velocity over two weeks is mostly him.
3. How he hits that hand. Left handed batters and right handed pitchers is a different matchup from left against left, and the gap is often larger than any edge you were looking for. Switch hitters get resolved to the side they will actually bat from.
4. The man on the mound. What he gives up, how deep he goes, and how he has handled bats from this side of the plate this season.
5. What that pitcher throws. His arsenal by pitch, split by the handedness of the batter, against how this batter has handled each of those pitches. This is the difference between “he is a good hitter facing a good pitcher” and “he has not touched a slider all year and this man throws one in three”.
6. The ballpark. A run environment, not a mood. Some parks add home runs and some remove them, and the effect is large enough to move a total bases line on its own.
7. Tonight’s air. Temperature, and the wind, taken as a direction relative to the way the park faces rather than as a number of miles per hour. A closed roof removes the question entirely.
8. The price. What the book is charging for the same prop, and what that price implies about how often it thinks the thing happens. A read that agrees with the market is not an edge, it is a confirmation.
9. The model’s number. One probability for one player clearing one line in one game, banded into a letter so a board can be scanned.
10. The record behind that number. What reads at that stated probability have actually done, from our own graded outcomes. Every probability the product shows you has been mapped through that table first, and the mapping is only allowed to lower a number.

The last two are the ones worth arguing about. A model probability is an estimate, not a measurement, and a letter on a card is a ranking, not a promise. What makes either of them worth reading is the row underneath: what reads like this one have actually done, counted, and published whether or not it helped.

What ET does for MLB

16 surfaces. Each one answers a question you would otherwise answer by opening five tabs.

What one of those cards actually holds

Two of those rows carry most of the work, and a sentence each undersells them. “Shows the reasoning” is a claim until somebody itemises it, so here is a hitter card, panel by panel.

Behind those panels is a glossary you never have to leave the card to read. 69 stats each carry a plain-English definition, the league average, where this player sits against it, and a line on what that stat does to each of the four batter markets. Tap the number and it tells you what it is. Hard-hit rate says it is the share of contact at 95 miles an hour or better, that the league runs 37.5%, and that a hitter well above it is a stronger home run and total bases signal than a hits signal. That last part is the bit most tools leave out, and it is the difference between a stat and a reason.

The pitcher card is built the same way: strikeout trends and the signals behind them, the season line, and what he allows to left and right handed bats with the weaker side named outright. Then home and away splits, his arsenal broken out by batter hand, the same game log with a line on it, and his last starts.

None of it tells you what to bet. It is assembled so you can disagree with the rating sitting on top of it, which is the only reason to show the working at all.

Which MLB markets it grades, and how much of each

Every read is written to the database with its probability, its line and its side before first pitch, so it cannot be edited afterwards. A read is settled against the official box score once the game is final, never against our own estimate of what happened. Here is the whole graded MLB record, read live from that table: 80,213 outcomes between April 14, 2026 and September 9, 2026.

Reads on players who never got into the game are voided rather than scored, and are not in these counts. A did-not-play is not a wrong read.

Strikeouts are counted here and published nowhere

The strikeout model ran about nine points hot. We rebuilt it, it went live on September 8, 2026, and we deleted the old model’s record rather than let it flatter or damn the new one. The count above is real graded outcomes. There is no strikeout accuracy table anywhere on this site right now, and there will not be one until the new model has enough graded reads to say something that is not noise. Anyone quoting one for us made it up.

What the record says, in one line each

The full tables live on the methodology page, banded by the probability we stated, with the sample size beside every row. Two of those rows belong here, because a page about baseball research that made you leave to find out whether the numbers hold up would be doing the industry’s trick.

• Where we say something is unlikely, we are close to right. Home run reads we put at about 7.2% landed 7.6%, across 3,356 graded outcomes. That band is the largest and steadiest part of the board.
• Where we say something is likely, we are too confident. Home run reads we put at about 50.0% landed 23.0%, across 631 graded outcomes. That is a gap of 27 percentage points, and it is the worst row we have.

The shape is the same in every market with enough data to check, and it is a known failure mode rather than a mystery. Rank thousands of candidates, keep the top ones, and the top of that list is partly genuine quality and partly luck that has not reverted yet. The luck reverts.

Read the ordering, not the number. A read we rate highly still lands more often than one we rate poorly. It just does not land as often as the number on it says, and the further up you look, the wider that gap gets.

Parks and weather, and what they can and cannot tell you

Baseball is the sport where the building matters. A fly ball that clears the wall in Denver is caught on the track in San Francisco, and the difference is large enough to move a total bases line by itself.

A park factor is a number that says how many more or fewer home runs, or runs, a ballpark produces than an average one. 1.30 means about thirty percent more. It is a multi-season average, so it describes the building and not tonight.

Tonight is what the weather is for. Wind is the piece that moves most, and the product treats it as a direction rather than a speed: fifteen miles per hour blowing across the field is not the same event as fifteen blowing out to center, so the wind bearing is resolved against the direction the park itself faces, and the result is split by whether the batter is left or right handed. A closed roof removes the question. Temperature does what you would guess, warm air carrying a ball a little further than cold, and it is a much smaller effect than the wind.

What none of it can do is make a bad matchup good. Environment is a thumb on the scale, not the scale. If the read only works because the wind is blowing out, the read does not work.

Where our park factors are weak

They are borrowed. Multi-season league-wide averages rather than numbers we recomputed from our own data, and they live in two places that are kept in step by hand: the constants the model scores against, and the table the app displays.

They were badly wrong once. Through the spring of 2026 the home run factors were inflated by half to double, with Dodger Stadium carrying 2.12 where the right number is about 1.00 and Yankee Stadium 1.88 where it is 1.15. Every home run read that crossed those parks was scored against a number that was not true. They were corrected in June 2026.

Which is why this page checks rather than reassures. When it last rebuilt, the two copies were compared across all 33 parks in the table, and they agreed on every one. Building our own from our own data would remove the hand-sync entirely. It is not built.

One prop is not a slip

Most people who research an MLB prop are about to put it on a ticket with three or four others. That changes the question, and it is where most of the money goes.

Legs in the same game are not independent

We measured this on our own data rather than repeating the folklore: every box score from the 2024 and 2025 regular seasons, 4,859 games, with a random cross-game pairing as a control that came back at 0.992 to 1.004, which is the number a null should look like. Both seasons had to agree in direction before anything was reported.

• A batter’s home run and his own team’s moneyline land together about 1.31 times as often as independence. It is the strongest pair in the data and it barely moved between seasons.
• His bats over and the opposing starter over fight each other, at 0.78 to 0.94 depending on the markets. That is exactly the slip somebody builds when they say they like both sides of a game.
• Two teammates both getting a hit is almost nothing, about 1.02. The shared offense that lifts total bases barely touches hits, which is why we price the bases pair and refuse to price the hits one.
• Two things we expected and did not find. A hot offense does not help its own starter strike people out, and home run weather lifting both teams did not survive the measurement. Both are reported here because we would have used them if they had worked.

How many legs a market can actually carry

A leg that hits one time in six is a one-leg bet, whatever the payout looks like. Once you can choose your own stake, the leg count that grows a bankroll fastest is fixed by the per-leg probability alone, and it is far shorter than what most tickets carry. On this board that works out at roughly four legs for hits, two for total bases or strikeouts, and one for a home run.

The product shows you that arithmetic beside the slip. It does not rebuild the slip for you, and it does not hide a lane it disapproves of.

And the book charges you for each one

We measured the sportsbook’s margin on 66,100 two-sided MLB prop prices. The cheapest MLB market to bet is pitcher strikeouts at 5.98%; the priciest is pitcher outs recorded at 7.02%. That is charged per leg, before anybody is right about anything. The whole table, every market and sport, is the Parlay Tax Index. A cheap market is not a good bet. It is a smaller head start for the house.

Why legs multiply, what a same-game parlay does to the price, and how to check a ticket before you place it are covered in why parlays lose, why same-game legs move together and how to check a slip before you bet it.

Where the numbers come from

Two different questions, and we answer them differently on purpose.

For grading, we are specific. A read is settled against the official box score once the game is final, never against our own copy of a result. If the record says a batter had two hits, the league says two hits, and you can check it in ten seconds. That is the part that decides whether anything above is worth reading, so that is the part we pin down.

For producing a read, we are not. The models run on public league data and a commercial odds feed. We do not itemise them, for the same reason we do not publish the model: naming the ingredients is most of the recipe. Treat that as a real limitation rather than a footnote. You can audit our outputs, not our method, and this page is built so the outputs are enough.

What we will not claim

Stated plainly so you can hold us to them:

• No single accuracy number. Pool every market together and you get a flattering figure that mostly measures something trivial, that home run reads are rarer than hit reads. We measured that it is an artifact and wrote a rule against it: no pooled performance metric, ever, on any surface.
• We do not claim to beat the sportsbook. We had one number on hits that looked like an edge. We ran the check on September 9, 2026, and it did not survive: prices are captured on a fifth of our graded reads and that fifth is heavily skewed toward reads we had already decided were likely. The working is on the methodology page, including the parts that make us look worse.
• We do not claim to beat the closing line. We do not store closing prices for MLB props, so we could not check that claim even if we wanted to make it.
• Pitcher outs reads are directional only. That model is the worst calibrated thing we ship, by a distance, and it is being rebuilt. It is on the board with that warning rather than quietly removed.
• We do not sell picks. ET produces reads. You choose every leg, and you place every bet somewhere else.

Common questions

Try it on a bet you were going to place anyway

The slip checker is free, needs no account, and reads MLB props: paste the legs the way your book prints them and it puts a letter on each one, tells you how often that letter has landed, and combines them honestly instead of multiplying and hoping. Nothing there is a pick. You bet wherever you already bet.

If you want the surfaces above, ET Plus covers MLB, and the record they are built on is on the methodology page before you pay for anything.

Nothing here is betting advice, and no research changes the fact that these are negative-expectation wagers. Bet only what you can afford to lose. If it stops being fun, take a break.