Brick by Brick

Congratulations. You bet on a baseball player hitting a baseball over a fence, which happens to any given hitter about once every eight or nine games, and he didn't do it today. You then did what every losing gambler in human history has done: you blamed literally everyone except the person holding your phone.

This book exists because the same thing happens every single week. The homers hit, everybody's a genius, everybody's screenshotting slips. Then the homers don't hit for two days, and suddenly the channel sounds like a hostage negotiation.

Here's the truth nobody wants tattooed on their forearm: the model didn't fail you, your bet sizing did. A system can be up 200 units and you can still be broke tailing it. That's not a paradox. That's the most common outcome in sports betting, and by the end of this book you'll see exactly how it happens, with charts, so you can't pretend you didn't.

Read that twice. The guy who builds the model has fucked this up. Everyone has. The difference between the people still here in a year and the people who rage-quit to go bet parlays on Bulgarian volleyball is one thing: they stopped changing their bet size based on how they feel.

A unit is a fixed dollar amount you bet on every play. That's it. Not "a unit, unless I really like it." Not "a unit, but double on Tuesdays because Mercury is in retrograde." A fixed number. When a capper says "+50u," it means if you bet $10 a unit you made $500, and if you bet $100 a unit you made $5,000. The units are how we talk about results without comparing wallets.

Your unit should be a small slice of a bankroll: money you've set aside for betting and could light on fire without missing rent, a car payment, or your kid's birthday. The usual guidance is 1–2% of bankroll per bet for normal stuff. For longshot homers, where you're firing ten-plus plays a night, go smaller, around 0.5–1%. And here's the rule that actually matters: keep everything you have in play on a night under about 5% of your bankroll, total. Ten plays? Each one is 0.5%. Fifty plays? Each one is 0.1%, which is $1 on a $1,000 roll, and if that's below the app minimum, you're playing too many, not betting too little. Chapter 5 shows why, with math that will make you uncomfortable.

Different products, different unit sizes

If you tail more than one thing, they don't all get the same dollar amount. The more plays a product fires per slate, the smaller each one should be, because the thing that kills you is total money at risk per day, not per bet.

So: if homers are $10, the 6-mans and 3-mans are about $5. Not $10. Definitely not $50 because "the 6s are up 150u." The 6-mans are up 150u compared to 80 for the 3s, and the EV really is higher, but higher EV with bigger swings needs a smaller bet, not a bigger one. We'll get to why.

Your unit is a promise you make to yourself when you're calm, so the idiot you become when you're tilted doesn't get a vote.

Let's do the math your brain skips when you're scrolling your slips at 11pm.

A home run at +500 pays 5-to-1. To break even you need to hit 1 in 6 (16.7%). A good model might hit 17–20% at that price. That means even when the model is crushing, 80% or more of your homer bets lose. You're not buying winners. You're buying a small edge on a bet that loses most of the time.

Here's the part that should be printed on every sportsbook: how often you'll hit a losing streak of a given length over just 100 bets (that's about ten slates of homers, not even two weeks).

Read the 16.7% row. A 10-bet losing streak is basically guaranteed (98%). A 15-bet losing streak happens 7 times out of 10. A 20-bet drought happens a third of the time. And this is a model that's breaking even, not a bad one.

So when you go 0-for-12 on a Tuesday and start typing "this shit is rigged," what you're actually saying is "I just experienced the single most predictable event in longshot betting and I'm shocked." That's like getting wet in the rain and demanding a refund from the sky.

A 0-for-15 isn't a signal. It's a Tuesday.

Here's a simplified version of how a high-variance system actually pays. It hits big one day, then bleeds three:

+50u, −10u, −10u, −10u  ·  +50u, −10u, −10u, −10u  ·  repeat

A high-IQ individual looks at that and sees +20u every four days. They also notice it loses three days out of four, and that the day right after a big win is the day it's most likely to lose. So they don't raise their units after a heater. They just keep laying bricks.

Now watch the casual gambler run the exact same picks. Same model. Same results. The only thing that changes is how much they bet, and they change it based on feelings.

Notice the Degen never picked a single bad bet. Every pick was the model's pick. They lost everything on bet sizing alone. That's the whole book, honestly. You could stop reading here. You won't, because you're about to see yourself in Chapter 4.

Every single stage is an emotional decision disguised as a strategic one. "It's house money." "I'll wait for a better spot." "I just need one day to get back." "Fuck it." None of those are strategies. They're moods with a bet slip attached.

And here's the cruel part: the degen and the bricklayer had the same edge. The degen didn't lose because he was unlucky. He lost because he took a positive-expectation system and bolted a random number generator made of emotions onto the bet size.

The four-day cycle is a cartoon. So let's do it properly. Below, 2,000 imaginary bettors each start with $1,000 and bet 10 homers a slate at +500 for 150 slates, basically a full season. Every one of them gets the same kind of edge. The only difference between the four groups is how they size bets:

• Half-Unit Hank bets $5 flat. Every bet. Forever. Boring as hell.
• Flat Freddy bets $10 flat. Still disciplined, just bigger.
• Press-It Pete raises his bet 50% after every winning slate and never lowers it. "Riding the heater."
• Chasing Chad doubles his bet after every losing slate to "get it back," then drops to $10 after a win.

Nobody can bet more than 10% of what's left in their account per bet. Drag the slider to change how good the model is. It starts at the top, 17.5%, which is a 5% edge. That's the best it gets; Chapter 7 explains why. At 16.7% there's no edge at all.

At the default setting (a 5% edge, the best case), look at what happens:

Pressing after wins goes broke every single time. Not most of the time. Every time. Because you raise your bet on the way up, you're betting your biggest amounts right before the inevitable drought, and one normal cold streak at 3x size erases everything the heater made. Chasing is almost as bad: it wins small most weeks, then one ugly losing run doubles you straight into zero.

And look at Freddy versus Hank. Freddy makes more money on average, but a real chunk of Freddys go broke too, and they went broke with a winning model. Ten $10 bets a slate on a $1,000 roll puts 10% of your bankroll in play every night. That's too hot for longshots. Hank almost never goes broke. This is why homer units should be baby-sized.

Now drag the slider down to 16.7% and below. With no edge, nothing saves you: discipline just loses slower. Units don't create an edge. They keep you alive long enough for a real one to show up.

Enough cartoons and fake seasons. This is the real thing: every SPOT Slugger home run play from to , graded at the listed price, 1 unit each. Every miss is in here. Every 0-for night is in here. Nothing's cherry-picked.

Play-by-play: pick a night

Here's every home run bet from any night, in the order it's graded, with the running total. Walk through the losing nights. Feel them. Then remember the whole thing finished way up.

Five guys tail the exact same bets

Same 28 nights, same plays, same results. Each guy starts with $1,000. The only difference is how they size bets:

• Half-Unit Hank uses the Chapter 18 calculator: 5% of his bankroll a night, split across every play, never more than $10 a bet. A 37-play night is $1.35 a play.
• Flat Freddy bets $10 a play, no matter how many plays. On a 37-play night that's $370 in action.
• Press-It Pete starts at $10 and raises 50% after every winning night. Never comes back down.
• Chasing Chad starts at $10, doubles after every losing night, and goes back to $10 after a win.
• Big Unit Barry bets a flat $50 a play. "Flat is disciplined, right?" Flat, yes. Disciplined, no.

Nobody can put more on a night than they have left.

Now shuffle the nights

You don't get to pick the order your nights come in. The heater could come first or last. So take the same 28 real nights and deal them in 2,000 random orders. Same bets, same total, different luck of the draw:

Honest fine print: this stretch was hot. It returned on the units risked. That's several times the ~5% edge that's realistic over the long run (next chapter). On a hot run, the big bettors look like geniuses whenever they survive. On a normal run they bust far more often; go back and look at Chapter 5. So no, smaller doesn't always make the most money in a given month. What it does is never go broke, in any order, on any run. That's the only way to still be here for the next heater.

You don't get to choose the order. So bet like the cold stretch comes first.

Time for the unsexy truth. No edge is guaranteed, and no real edge on the book is over about 5%. Not ours, not anybody's. The book prices every homer with its cut already baked in, so the best models in the world are fighting for a few cents on the dollar. Anyone showing you a 30% ROI is showing you a hot month, a cherry-picked screenshot, or a lie.

So what does a 5% edge actually get you? On $10 bets, $0.50 of expected profit per bet. That's it. A hundred bets is about ten slates of homers, and that's worth about +$50 on average. The swings around that $50 are ±$230. Read those two numbers again: the noise is almost five times bigger than the edge.

That's with a perfect 5% edge. In real life you don't even know your edge exactly. It moves: the market wises up, rosters change in September, the wind flips. A system that ran +5% for three months can run +1% for the next three, and at +1% the chart says you're basically flipping a coin all season.

This is why every rule in this book is about surviving, not about maximizing. Discipline doesn't make you a winner. It keeps you in the game long enough to find out whether you are one, and it keeps the losing seasons from wrecking your life.

Nothing is guaranteed. Not the model, not the heater, not tomorrow's slate. Size every bet like that's true, because it is.

Every price the book hangs has its cut baked in. Add up the implied chances of the Over and the Under and you get more than 100%. That extra is the vig (also called juice or hold), and it's how the book gets paid whether your guy homers or not. De-vigging strips that cut out to estimate the "fair" odds: what the market thinks the real chance is.

It's a useful tool. It's also the most misused number in betting Twitter. First the math, then the lie.

The math, line by line

Step 1: Turn each price into an implied probability.

Step 2: Add the Over and the Under together. In a world with no vig it'd be exactly 100%. It never is.

Step 3: Find the book's cut (the hold).

Step 4: Divide each side by the total. Now the two sides add to exactly 100%. These are the fair (no-vig) probabilities.

Step 5: Turn a fair probability back into odds.

Step 6: The only step that makes you money. Compare your probability to the price you're actually getting, not to the fair price:

That's it. Now watch it done on the board. Type in any market below and both boards redo the math live.

Convert any odds, any direction

Books, exchanges and apps all speak different languages. American (+350), decimal (4.50), fractional (7/2) and implied chance (22.2%) are the same price written four ways. Type into any box and the other three follow, then watch the board do the math.

Now the lie: fair odds are not your edge

Here's where people go broke feeling smart. They de-vig a market, see "fair +375" next to a book offering +400, and scream +EV!! Sometimes it is. Often it isn't. Here's what the de-vig does not tell you:

1. "Fair" depends on how you de-vig. There's no single correct way to split the book's cut. The multiplicative method above splits it evenly by size. But books usually load more of their cut onto the longshot side, because that's what the public bets. Other methods account for that and give the longshot a lower fair chance. Same market, different "fair" answers:

2. You don't get paid at the fair price. You get paid at the book's price, vig and all. So your probability has to beat the book's implied number, not the fair one. If fair says 21% and the book's price implies 22.2%, a model that says 21.5% "beats the fair line" and still loses money. That gap is the vig you're paying. Compare your number to the price you're clicking, always.

3. The market isn't the truth either. De-vigging gives you the market's opinion with the cut removed. If you only bet when you beat the de-vigged market, you're betting that the market is wrong, and it usually isn't. Your edge has to come from somewhere real: a model that's actually calibrated, information the price hasn't caught up to, a soft book lagging a sharp one.

4. One weird price is usually a mistake, not a gift. A single book way off everyone else is often a stale line, a bad quote, or a limit trap. The bigger the "edge" looks, the more likely something is broken. Check the price against the rest of the market before you get excited.

5. Lots of props are one-sided. Homer props often only list the Over. No Under means nothing to de-vig, so you're guessing the hold. On longshots it's usually bigger than you'd think.

6. Your model's number has error too. "My model says 24%" really means "somewhere around 20–28%." On a 22% event, a 2-point edge is inside the noise. Remember Chapter 7: 5% is the ceiling. If your math says 25% EV on something, your math is wrong, not the book.

7. The real report card is the closing line. If the prices you bet keep moving your way before the game (you took +400, it closed +320), you're probably beating the market. If you're "+EV" on paper but the lines keep drifting away from you, the market is telling you something.

PrizePicks-style apps don't show odds at all, so there's nothing to de-vig. They get their own chapter, next.

De-vig tells you what the market thinks. Only a real edge over the price you're actually paying makes you money.

Pick'em apps are sneaky. There's no −110 staring at you, no odds at all. Just "3x!" and "37.5x!" in big happy letters. So people assume there's no juice. Wrong. The juice is hidden inside the multiplier, and on some entries it's way worse than any sportsbook.

The move is to turn every payout into one number: the hit rate each leg needs just to break even. If your picks don't clear that, you're donating.

The math, line by line

Power plays (all legs must hit). You get paid when every leg hits, so:

2-pick Power pays 3x → (1 ÷ 3)^(1 ÷ 2) = 57.7% per leg. Compare that to a normal sportsbook bet at −110, which needs 52.4%. The 2-man charges you like a −137 line on every single leg.

Flex plays (partial payouts). These need the binomial formula: the chance of hitting exactly k of n legs is C(n,k) × p^k × (1−p)^(n−k). Multiply each by its payout and add them up. Set that equal to 1 and solve. The fun part: do the algebra on a 3-pick Flex (3x for 3/3, 1x for 2/3):

A 3-pick Flex is a 2-pick Power in disguise: identical math, identical 57.7% break-even. Same trick on a 4-pick Flex (6x / 1.5x) gives 6p³, which is exactly a 3-pick Power. The "safety net" doesn't change your expected value at all. It just spreads the same payout over more outcomes.

Why your 2-man and 4-man are a tax

Look at the chart again. The 2-pick Power (and its twin, the 3-pick Flex) needs 57.7% per leg: the worst price on the whole app. Hitting 57.7% on player props long term is elite. Most people who think they're hitting it are remembering their wins and forgetting their losses. If you're firing 2-mans every night because "it's only two legs, it's basically free money," you're paying the highest juice PrizePicks charges and calling it the safe play.

The 4-pick Power is the second-worst Power entry, at 56.2%. It pays 10x. To be as good a deal as the 3-pick Power (6x), it would need to pay about 10.9x. So you're taking on an extra leg of risk and getting paid less per leg for it. There's no reason to play a 4-man Power over a 3-man Power, and anyone doing it doesn't know that, because they've never done the math.

The best prices on the app are the 5- and 6-pick Flex (about 54.2%) and the 5- and 6-pick Power (about 54.7–54.9%). That's why the serious guys play 6-mans. Not because 37.5x looks sexy on a screenshot, but because it's the cheapest juice.

Compounding cuts both ways

Here's the catch that makes 6-mans dangerous for degens. A small edge per leg compounds across six legs, so the entry EV looks huge: at 57% per leg a 6-pick Power is about +29%. But a small mistake per leg compounds the same way: at 53% it's about −17%. Four points of per-leg accuracy is the difference between a great entry and lighting money on fire. If you can't tell whether your picks hit 53% or 57%, you have no business guessing.

And the variance is brutal. At 57% per leg you hit all six about 3.4% of the time: roughly 1 entry in 29. There's about a 17% chance of going 50 straight 6-pick Powers without a 6/6. That's exactly why the 6-mans get baby units, half your homer unit or less, and why they got their own channel. It's the four-day cycle from Chapter 3 on steroids.

Nobody beats the books by vibes

Say it with me: nobody beats the books who isn't crunching the numbers. Not your cousin who "just knows ball." Not the guy in the Discord with a 9-leg screenshot from April. The app is built by people with PhDs who've done every calculation on this page and a thousand more. If you haven't done them, you're not playing against the app. You're funding it.

The bare minimum before you enter anything:

1. Know the break-even of the entry you're playing (the chart above).
2. Know your real per-leg hit rate from a tracked record, not memory.
3. Only play entries where your rate clears the break-even by a real margin, not by a hair.
4. Size by the variance: the more legs, the smaller the entry.
5. Check the payout on the screen every time. Multipliers change, goblins and demons pay less or more, and some states differ.

A 2-man isn't the safe play. It's the most expensive play on the app with a friendly font.

Here's the chapter that separates people who make money from people who "had the right pick." The same bet at two different books is two different bets. One can be +EV and the other a donation, on the same player and the same line.

Quick math. A normal side at −110 needs 52.4% to break even. At −105 it needs 51.2%. At +100 it needs 50.0%. Your whole edge is usually 1–3 points. So a lazy −110 instead of an available +100 can eat all of it, and you'll never know, because you still "hit your picks."

Real Oracle bets, three ways

Oracle posts a play with every book's price attached. So we can replay every posted Oracle bet three ways: at the best price on the board, at the typical (median) book, and at the worst book. Same picks, same results. Only books you can actually bet count, and junk quotes are removed.

At the best price: +3.4 units. At the typical book: −47.0 units. At the worst book: −76.2 units. Same 1,025 bets. The picks didn't change. The price did, and it was the difference between breaking even and getting cooked.

Look at how thin the top line is, too. That's the honest truth about +EV betting: the edge is small, and it only exists at the best number. Take the median book and you're not "a little less profitable." You're a losing bettor.

What that looks like, one bet at a time

Notice the losses cost 1 unit no matter where you bet. Only the wins change. Take the worse price and you pay full price for every loss while getting paid less on every win. Over a thousand bets, that's the entire gap between the three lines above.

Shop in 60 seconds

1. Have accounts at several books. Three is the minimum and five is better. More books means a better best price.
2. Check every book before you click. Use the free line shopper in Chapter 14 or the price table on every Oracle card.
3. Know your "no worse than" price. If your book is below the posted price, check the EV at your price (the Oracle guide shows how). If it's negative, pass.
4. Distrust a price that's way better than everyone else's. In NFL Week 1, ten yardage props showed up at +900 to +2000 when every other book had them near even. They were bad quotes, not gifts, and they would have faked a +44% ROI. A real edge is a few cents, not a few hundred.
5. Missed the number? Skip it. There's another bet tomorrow. There are 100,000,000 games left.

You don't have an edge. The price has an edge. Go get the price.

Oracle is the +EV bot. It scans every book, strips the vig out of the market to get a fair price (the Chapter 8 math), and pings the Discord when one book is paying more than fair. It doesn't guess winners. It finds prices that are too generous. You win by taking those prices over and over.

Reading a ping

Illustration. The fair odds, the best price (NoVig +117) and the worst price (Fanatics −110) come from the real Sep 10 play. The middle two rows are filled in to show the layout.

Take

The bet and the best price Oracle found. That price is the whole reason it pinged.

Fair odds

What the bet is "really" worth after stripping the vig from every book (the median). Anything paying more than fair is +EV on paper.

EV

Expected profit per $1 at that price. Oracle only pings roughly +3% to +5%. Bigger "edges" are almost always bad quotes, so they get filtered out.

Book table

Every book's price and its EV. Go down the list until you find one you can bet with a positive EV. If none of yours are positive, you pass. Seriously.

The five rules

1. Take the longest price you can get. Chapter 10 showed it: best price about breakeven-plus, median book −4.6% a bet. Oracle's edge is the price.
2. Never chase a moved line. If the price dropped below fair by the time you see it, the bet is gone. Don't take it anyway.
3. Flat 1 unit, every ping. Oracle's edges are small and similar in size. Don't size up the one you "like."
4. Mind the volume. Oracle can fire dozens of pings a day. Keep the whole day under your nightly cap (Chapter 18). Lots of pings means a smaller unit.
5. Judge it on hundreds of bets and on price, not on wins. Plenty of these are plus-money, so it loses more bets than it wins. The question is only whether you're consistently getting better than fair.

Check your own price

Got a worse number than the ping? Plug it in. If the EV comes out negative, skip it.

The honest record: across those 1,025 posted plays, Oracle was about breakeven at the best price and clearly negative at the typical book. It's a price-finding tool, not a money printer. It only works if you do the shopping part.

Slugger is the MLB model. For every hitter it builds a projection from his own numbers, adjusted for the pitcher he's facing and the park, and turns that into a chance of hitting the prop. Then it compares that chance to the market's de-vigged fair price. When the model says "more likely than the market thinks," at a price that pays enough, it lands on the card.

Reading the box

Every hitter gets a box for every prop. Here's a real home run box from the board, piece by piece:

Recreated from the live Slugger board (Sep 23). The numbers are the real ones on the board.

1. The hitter. His name, which side he hits from (RHB = bats right), and how many games his numbers are built on (108g).
2. Projection. How many homers Slugger expects tonight: 0.36. Built from his own power numbers, then adjusted for tonight's pitcher and park.
3. Edge. Slugger's chance minus the market's de-vigged chance, in percentage points. +11.2 means Slugger thinks he's about 11 points more likely to go deep than the market does.
4. The line. o/u 0.5 homers, so one homer cashes the Over.
5. The badge. SPOT means it's a play. SPOT+ is the same bet with a bigger EV; bet it the same size (see the tiers below). An outlined THIN SPOT / HELD SPOT cleared the numbers but is withheld on purpose. Don't bet those.
6. The price, lit up. OV +470 is the best Over price across every book, and it's bright because it's the side Slugger likes. If your book is lower, the value shrinks fast.
7. EV. +70% is the expected profit per $1 at +470. Homer EVs look huge because a ~20% shot times a +470 payout swings a lot. It's not a promise, and it disappears if you take a worse price.
8. What the price implies. +470 implies 18%, the raw chance that price claims, vig included. Slugger's number is higher; that gap is the whole bet.
9. The other side, dimmed. UN −506 (83%) is shown for reference. Dim means not the play.

Also on the board: ■ means the price is locked (frozen pre-game). A symbol next to a price († thin sample, ∅ over +1200, ≫ skew, ½ one-sided market) means the lean is withheld. No badge, no bet.

The tiers

How to bet the card, step by step

1. Count tonight's SPOT plays. Some nights it's 3 and some nights it's 37.
2. Set tonight's unit: your nightly cap (5% of bankroll) ÷ number of plays, and never more than your per-bet max. The Chapter 18 calculator does it for you. 37 plays on a $1,000 roll is about $1.35 a play. That's not a typo either.
3. Shop every play. Bet each one at the best price you can get. If your book is well below the listed odds, the EV shrinks fast on longshots. +700 vs +550 is a massive difference in what a hit pays.
4. Skip the ∅ plays (over +1200) unless you enjoy donating.
5. Don't add your own "locks" on top, and don't cut the ones you don't like. The card is a portfolio. Cherry-picking is how you end up skipping the 3 that homer.
6. Walk away. 84% of these lose. You already knew that (Chapter 2). Check results in the morning, not pitch by pitch.

Tail the whole card, size it by the night, shop every price, sleep.

Here's every Slugger prop that has made money, graded from 08/21 to 09/22 (33 nights), each at the threshold where it's held up best. Every threshold only sets a floor (EV at least X, edge at least Y, price no longer than Z), never a hand-picked window. To count, a threshold needs 60+ bets, has to be up units, green on at least half the nights it bet, and up in both halves of the stretch. Flat 1 unit, at the best price. Bad quotes (prices that contradict their own ladder) are thrown out before anything is counted.

Not on the card: Earned runs (pitcher) over, Earned runs (pitcher) under, Home runs under, Hits+Runs+RBI over, RBIs over, Runs over, Runs under, Strikeouts (pitcher) under, Walks under. No threshold on these held up across the stretch, so we don't bet them, and they aren't here.

Read this before you bet a single threshold

These thresholds were picked by looking at these same 33 nights. That's the honest weakness of any "best threshold" list: if you test enough cutoffs, some look great by luck. We checked. Shuffling which bets won and re-running the whole search still turns up about 12 props that "pass", against 15 on the real results. So a few cards above are probably luck, and we can't tell you which. Expect every number here to come down going forward, and trust the ones with the most bets and the steadiest second half.

A model can be profitable without hitting more at the top. Slugger's highest-confidence plays don't always hit more often than its middle ones. There are things it doesn't account for (late scratches, weather shifts, bullpen days), and it's weakest where data is thin: call-ups, rare matchups, short samples. That's why these thresholds are floors and not "bet more on the biggest number", and why SPOT+ is bet the same size as SPOT.

Bet the floors, size the night, shop the price, and judge it on hundreds of bets.

The Kelly criterion is the math answer to "how much should I bet?" You give it your chance of winning and the price, and it gives you the share of your bankroll that grows money fastest over time. Bigger edge, bigger bet. No edge, no bet. Sounds perfect.

The catch is in the p. Kelly assumes your win chance is exactly right. If your model is even a little overconfident, Kelly doesn't bet a little too much. It bets way too much, over and over, and the bankroll pays for it. That's why pros bet a fraction of Kelly: half, quarter, or less.

Why the fraction matters: a model that's a little too sure

Here's the trap in one example. A model says its longshots hit 21% of the time at +500. They really hit 17.5%. That's still a winning model: +500 only needs 16.7%, so it has a real edge. But Kelly sizes every bet off the 21%, so it bets like the edge is four times bigger than it is. Below, 1,000 bettors each run 150 nights of 10 of those bets from $1,000. The only difference between them is the Kelly fraction.

This is common, and it's not a knock on a model. Plenty of models, profitable ones included, don't hit more often at the top of their confidence scale, because there are things they aren't accounting for (a late scratch, a wind shift, a bullpen game) and they're shakiest where they have the least data: new call-ups, rare matchups, weird parks. Their 30% plays don't always hit more than their 20% plays. They can still beat the price, but you can't size bets on their exact number.

The Kelly rules

1. Never full Kelly. Not on props, not on anything. It assumes a perfect probability, and yours isn't.
2. Quarter Kelly is the ceiling; eighth for longshots. You give up a little growth and cut the crashes way down.
3. Kelly per bet, cap per night. Twenty "small" Kelly bets on one slate is one big bet. Keep the night under your nightly cap.
4. No probability, no Kelly. If you're tailing without a real win chance (a tracked, calibrated number), stick to flat units.
5. Check the model against reality. If the plays hit less often than the model says, you'll overbet on every single one until you cut the fraction.

Kelly is a scalpel. Full Kelly on an overconfident model is a chainsaw.

A drawdown is how far you fall from your highest point before you climb back. Every winning system has them. Every one. The only question is how deep.

Think of drawdowns as rent on the edge. You don't get the +200u season without also living through the −60u stretch in the middle of it. The people who make money are the people who expected the −60u, sized so it didn't kill them, and were still tailing when the next +50u day showed up.

What to actually do in a drawdown

Nothing. That's the move. Same unit, same plays, same process. The drawdown is the model doing exactly what the math said it would. The only reasons to change anything are a real change in the process (the model changed, the market changed, the season's ending with weird rosters) or a change in your life (you need that money now). "I'm down and I'm pissed" is neither.

If a drawdown is causing you actual stress (you're checking the app at work, you're snapping at people), your units were too big. Not the model. Lower them for next time and let the math do its thing.

Y'all forget we have 100,000,000 games left in our lifetime. Stop rushing. Start building.

1. Thou shalt pick a unit while sober and calmDecide it on a day you didn't bet. Write it down. That number is law until your bankroll moves a lot, up or down.
2. Thou shalt not raise units after a heaterA +50u day is the start of the cold stretch, statistically. Your ego says "press." The math says "same." Listen to the math, it's never been dumped.
3. Thou shalt not chase"Getting it back" is how people lose three times what they were down. The next slate doesn't know you lost yesterday. It doesn't care about your feelings.
4. Thou shalt not go light on "bad feeling" daysYou can't tell which day hits. Neither can I. If you skip the days you're scared of, you'll skip the +50u day. Guaranteed. It's a law of the universe.
5. If you raise, raise like a cowardA prime slate with the wind blowing out everywhere? Maybe go from $20 to $21. That's the kind of raise we're talking. Not 2x. Never 2x.
6. More plays per slate means a smaller unitHomers $10? The 3-mans and 6-mans get about $5. Total money at risk per night is what kills you, not the size of one bet. Keep the whole night under about 5% of your bankroll.
7. Thou shalt keep the bankroll separateIts own account, its own money. Rent money isn't bankroll. Credit cards aren't bankroll. "I'll pay it back Friday" isn't bankroll.
8. Thou shalt judge on 100+ bets, not one dayOne slate tells you nothing. One week of homers tells you almost nothing. If you can't wait a month to judge a system, you can't afford to bet it.
9. Thou shalt never go all inNot on a lock. Not on a "can't miss." Not after a bad day. All-in is how the Degen's chart ends every time. There is no chapter after all-in.
10. Thou shalt own thy resultsYou picked the size. You clicked the button. The model makes picks; you make bets. If you're up 10u when the model's up 40u, that 30u gap is yours.

Everything above is about playing a long game with a real edge. Some honest fine print, because you deserve it:

No system is guaranteed. Past units don't promise future units. Edges shrink, markets adjust, seasons end. Good bet sizing is what lets you find out whether an edge is real without going broke in the process. It doesn't manufacture one.

The best realistic edge is around 5%. Even at that ceiling, about 1 in 5 full seasons still finishes in the red (Chapter 7). Any capper, including this one, can have a losing month, a losing season, or an edge that quietly disappears.

Most sports bettors lose over time. The books take a cut on every bet. Anyone telling you betting is a guaranteed side income is selling something.

If you ticked items 6 or 7 in the mirror test, or you're betting to fix money problems, stop reading jokes and read this: