Skip to content
Enter the Lab →
The Lab

About ET Parlays

8:51 in the morning, Thursday, July 31, 2025. I’m making a two-page PDF about four hitters in a Rays-Yankees game.

Splits against righties. A five-game hitting streak. A note that the wind is blowing in at Yankee Stadium at 12 miles an hour. The hashtag #ETParlays, which at that point was the name of a PDF. And at the very bottom, where I forgot to delete it, a takeout order: fried rice or ice cream or cream cheese rangoons.

By September it was nine pages. A hits watchlist, total bases, home runs. A column called “Why We Like Him.” Ohtani, 1+ hit, 89%. Judge, 82%. Then the parlays I built out of them: a 3-leg, a 4-leg, a 7-leg I labelled “Aggressive,” and a 7-leg lotto.

I sent them to a couple of friends and most of the people I worked with at the hospital who watched sports and bet on them. Builder Thursdays. I knew there was variance. That’s why they were lists. People could make their own decisions, or their own parlay.

By April the PDF was an app, the lists were graded, and my own landing page said 61%.

April 22, 2026. Late afternoon, home, running audits on the landing page. I checked the number against the database.

764 graded. 36.9%.

The first thing I thought wasn’t “hide it.” It was: how do we improve that, or is the sample just small?

I didn’t lower the number. I deleted the whole bar and shipped nothing in its place. The site went four months without a performance stat on it, because the only honest number I had was one I hadn’t earned yet.

A week earlier I’d deleted the button that built the parlay for you. The thing I used to do by hand on Thursdays.

I’d built a machine to check the book’s numbers. The first number it caught was mine.

What I found when I checked

Since then, the MLB batter reads that make the board get graded after the game, win or lose. Here is the record. It’s read from the database when this page loads, so it can’t drift the way the 61% did.

11,692 reads graded between April 14, 2026 and September 27, 2026. MLB batter props that made the board, each one written before its game started. One season, not ten.

One market at a time, because a home run and a hit are different bets with different odds. Each row is what the model said, how many reads it said it on, and how often they landed.

Home runs · 2,893 reads

Model saidReadsAverageLanded
10 to 20%35317%19%
20 to 30%74926%17%
30 to 40%1,14534%18%
40 to 50%43744%22%
50 to 60%20750%26%

Total bases · 3,055 reads

Model saidReadsAverageLanded
30 to 40%38338%41%
40 to 50%2,39547%41%
50 to 60%19553%37%
60 to 70%5667%29%

One or more hits · 2,892 reads

Model saidReadsAverageLanded
60 to 70%1,45966%61%
70 to 80%1,17674%65%
80 to 90%24782%64%

Hits plus runs plus RBIs · 2,852 reads

Model saidReadsAverageLanded
40 to 50%34049%49%
50 to 60%1,73655%52%
60 to 70%73563%46%
70 to 80%4172%61%

Average is what the model said on average inside each band. Bands with fewer than 30 graded reads aren’t shown, because a number built on twelve outcomes isn’t a number.

Strikeout props are left out until their number is on the same basis as the rest.

The widest gap was on home runs. Before September 10, 2026, when the model said 50 to 60%, those reads hit 26%. On September 10 we cut the home run numbers the board shows. Since then, 329 home run reads said 18% on average, and 18% landed. That’s two and a half weeks of games, so it’s a start, not proof.

The postseason, counted on its own

From late September the board ran on one to four games a day instead of fifteen, and it still filled the same number of spots. So the top of the board was most of a few rosters, and pitchers and hitters with no batting data sat there on the model’s default numbers. Those reads said more than landed.

On October 13 we changed the board to score only hitters with a real batting line this season. The reads below were shown to people, so they are in the record, in their own table, instead of folded into the regular season or left out.

1,197 postseason reads graded between September 29, 2026 and October 28, 2026, each written before its game started.

MarketReadsSaid, on averageLanded
Home runs32212%7%
Total bases36234%27%
One or more hits23858%50%
Hits plus runs plus RBIs27543%39%

One row per market, because a postseason market has too few reads to split into bands.

Parlays were worse. From late May to September 8, 2026, the parlay lanes built 368 slips in internal testing. 129 of them were saved after one of their games had already started, so I took them out. Of the 234 left, we expected about 70 winners and got 17. The legs inside them said 71% and hit 46%. The combined odds we displayed were not achievable. An earlier version of this page said 348 parlays and 25 winners, late slips included. The late ones did no better, so the story held, but the count was wrong.

On August 24 I started cutting the MLB parlay leg numbers down by how much each prop type had actually missed, before multiplying them. Home run legs get cut the most. Then a same-game correlation adjustment goes on top, and the face says so. I later found that some of the slips I measured those cuts on had been saved after their games started. The home run and total bases cuts hold without them. The hits cut doesn’t have enough real predictions behind it yet. Whether the August fix worked overall is not graded yet. Most of those 234 slips predate it.

Slip Check now corrects each leg by how often numbers like it have landed before it multiplies them, then adjusts for legs we’ve measured moving together. It’s still an estimate, not a promise.

The full version of all of this, every denominator, how a read is graded, how much of the board never gets graded, and each place our numbers said more than landed, is on the methodology page.

I’m putting this on the About page because it’s the only proof of method you can check yourself. The full record is in the Lab. It includes the misses.

Things I tried that didn’t work

Each one had a test set in advance and failed it.

  • A hot-streak boost. May 3. Replayed the 2024 season, 38,054 batter-games. Hitters on a hot two weeks hit about 7 points below what the model already said. Widening the streak effect made every band worse. Reverted.
  • An NBA totals model. May 9. Replayed 1,236 games. It under-projected high-scoring games by 8 to 21 points. The totals number is still held back from the cards.
  • A rule to suppress KBO reads the model liked but the market didn’t. August 20. This one hurt. A friend used those reads every day, and I kept chasing an answer to the variance. When he texted me the KBO leans were off, I had to see the numbers myself before I could respond. They were eye-opening. The KBO reads came down that day. Then the replay killed my fix too: it would have deleted the only segment that made money at its price and kept the one that lost. Nothing shipped, and I had to tell him so.
  • Searching my own data for a hidden edge. September 2. 201 slices. The best result was inside the range pure chance produces. Nothing found. That one is on the list on purpose.

What the Lab does

You bet wherever you already bet. ET is the Lab you open first.

  • Every read comes with the reasoning shown, and every stat explained next to the number, in plain words. This is the thing members mention most.
  • Paste any slip and it checks every leg before you place it, and names the one dragging the slip down.
  • Every read that makes the board is graded after the game, and the record is on this page and in the Lab, misses included, so you can check ET instead of following ET.

A grade is a read, not a promise. Sometimes the best bet is no bet at all.

What it refuses to do

ET isn’t a sportsbook. It takes no wagers, holds no money, and takes no cut of anything you bet.

Options, not parlays. Nothing here is picked, staked, or placed for you. A button that bets for you is a button that thinks for you, and that’s how you stay the fish.

The subscription is flat. ET makes the same money whether you bet once this month or a hundred times. Almost nothing else in this industry can say that.

Straight up: no sportsbook pays ET anything today. If that ever changes, it gets written here first, and it won’t change the numbers we show you. What no book can buy is a grade, a ranking, or a spot on the board.

Who it’s for

If you go back and look at the slip after it loses, this is for you. If you’ve ever asked “was that a bad beat, or a bad bet?” and not liked that you couldn’t answer, this is for you.

If you want to be told what to bet, it isn’t. Plenty of people will do that. None of them will show you their misses.

Who’s writing this

I’m Eddie. I’m a nurse on a surgical progressive care unit. The monitor room calls: your patient’s in the 140s. I’m rushing down the hall, and when I get there they’re just chillin in bed. Happens all the time. The number is where you start, not where you stop. You look at the person.

Long before I ever wrote code, I wanted to build things. I started on AutoCAD in 2011, in a drafting class in high school. For years after that I went back and forth between architecture and mechanical engineering. In college I worked in an engineering shop as a machinist, using software to cut parts on the machines. Then the pandemic hit, and I switched to nursing because I wanted to help.

I’ve been a sports fan my whole life, and for fun I’d bet parlays. I used the prop tools out there, but research was eating too much of my mornings, and it always left me with the same questions. Why did one leg ruin an otherwise good parlay? How reliable were the probabilities? What did the historical data actually support?

I wanted something I could use myself that gave me what I needed, and I wanted a challenge. By the summer of 2025 I had my own models. They ranked the players in the spreadsheets and PDFs I sent to coworkers. What started as a tool for me became something I shared at work, and then this. I’ve built the rest myself: the app, the jobs that grade the reads after each game, and this page, which pulls the record straight from the database. What I didn’t know, I learned from books on statistics and databases. And I’m still learning.

I’m still a full-time nurse. Most weeks I work three twelve-hour shifts, Friday to Sunday, which is when the games are, so most of this gets built on weeknights. In April 2026 a friend started texting me their slips so I’d check the legs by hand. That’s where Slip Check came from. The first version was me.

I don’t have a background as a professional sports bettor or a career in a Silicon Valley startup. I’m not going to pretend I know what’s going to happen tonight. I can show you what the number says, what it’s said before, and how often that turned out to be true.

And when the evidence tells me something I’ve built is wrong, I want to know.

That’s the story. Now check the method.

See where it’s been wrong.