How would you even know if you had an edge?
Winning money is not proof you are good at poker. A look at how much evidence it actually takes to tell skill from luck (in poker, in betting markets, and in baseball), and why most players never find out.
I run a bot that bets on sports markets. It's been up since the spring. My honest reaction to that wasn't “great, it works.” It was “I still don't know if it works.” A few months of profit is exactly what a strategy with no edge looks like a good chunk of the time. That gap, between I'm winning and I have an edge, is the most expensive misunderstanding in every game with money in it. Poker included.
So here's the question I want to sit with, because almost nobody actually answers it before they've risked a lot of money on the assumption: how would you even know if you had an edge?
The uncomfortable starting point
Poker between players is zero-sum: for every dollar won, a dollar is lost. But the house takes a cut of almost every pot. Once you subtract that rake, it's negative-sum. The money leaving the table is real and constant, which means the average player is a long-term loser, and not by a little.
I play poker, and I'm building a poker trainer. Which makes the next sentence awkward to write: in any game with a rake, most players are lifetime losers, including plenty who are convinced they're winners. I'd rather say that up front than pretend otherwise, because the whole point of getting better is to be in the minority the rake doesn't sink, and you can't aim at that if you won't look at it.
Why your results can't answer the question (for a long time)
The instinct is: check the bankroll. Up money, must be good. The problem is that poker results are drowning in variance. A perfectly good winning player will have losing months. A losing player will have winning ones, for longer than you'd believe.
Say you truly win at 5 bb/100 (five big blinds per hundred hands, a strong and realistic win rate). Because the swing in any stretch of hands is enormous, your measured win rate wanders inside a wide band that only tightens with volume. Here's that band, assuming a typical online-6-max swing:
Read the shaded band as: “given your real skill, here's where your results could honestly land.” For the first tens of thousands of hands it comfortably includes zero, meaning your results are consistent both with being a genuine winner and with being a breakeven player on a heater. You cannot tell which one you are. It takes roughly 150,000 hands before the band even clears breakeven with confidence.
“150,000 hands” is longer than it sounds
| Where you play | Hands / hour | Time to 150,000 hands |
|---|---|---|
| Live $1/$2 | ~30 | ~5,000 hours (years of regular play) |
| Online, one table | ~80 | ~1,900 hours |
| Online, multi-tabling | ~400 | ~375 hours (a few months) |
If you're a live $1/$2 player, confirming you beat the game takes years, and by the time you have the sample, you're a different (hopefully better) player than the one who generated it. This is not a footnote. It's the central fact of trying to know whether you're good.
One honest caveat: that 150,000 assumes an online-style swing. Live games swing a bit less per hand, so they need somewhat fewer hands to confirm an edge, but they deal so much slower that the clock time stays measured in years either way.
Same skill, five different lives
To make it concrete, here are five players with identical skill, all genuine 5 bb/100 winners, simulated over 100,000 hands each:
Same edge. Wildly different stories. One of them plunges thousands of big blinds underwater and stays stuck below breakeven for tens of thousands of hands, long enough to quit, switch styles, or decide the game is rigged, before finally climbing out. He was a winner the whole time; the sample just made him wait. When someone tells you their results prove they're a winner (or a loser), this chart is why I don't believe them yet.
This was never really about poker
The math above is a special case of a law that shows up everywhere you try to separate skill from luck: the number of observations you need scales with the variance divided by the size of your edge, squared. Small edge in a noisy game? You need an enormous sample. It's the same problem wearing different costumes.
Baseball ran into this decades ago and did the work of measuring it. Different stats become trustworthy at wildly different sample sizes:
| Metric | Roughly trustworthy after | Reading |
|---|---|---|
| Strikeout rate | ~60 plate appearances | believable fast |
| Batting average on balls in play | ~800 balls in play | mostly noise early |
| Poker win rate (winner vs. breakeven) | ~150,000 hands | believable very slowly |
These thresholds aren't measured the same way. The baseball figures mark where a stat becomes about half signal (the standard sabermetric “stabilization” point); the poker figure is where a true 5 bb/100 winner's results clear breakeven. Different yardsticks, same lesson: how much data you need before a number means anything swings wildly from one metric to the next.
A hitter's early-season batting average tells you almost nothing; his strikeout rate tells you a lot. Your poker win rate lives at the far, slow end of that spectrum. Knowing where a number sits on this scale before you trust it is most of the skill.
The shortcut poker doesn't have
Here's where my betting bot gets interesting, and where poker gets unlucky. In efficient betting markets there's a faster proxy for edge than profit: closing-line value. The closing line, the final odds right before an event, is the market's best guess, and it's hard to beat. If you consistently get a better price than the close, that's strong evidence you have an edge long before your profit-and-loss could ever prove it. It's a measurement of decision quality that doesn't have to wait for the results to arrive.
Poker has no closing line. There's no efficient market stamping a “correct price” on your river call. So you're stuck with the slow instrument, raw results, which is exactly why the poker version of this question is so brutal to answer.
So: do you have an edge?
The honest framework is three questions, in order:
What you can actually measure
If results are a slow, noisy instrument, grade the thing you control instead: decision quality. Did you make the play with the best expected value, given what you knew? That you can evaluate one hand at a time, with feedback in seconds instead of years, and it's the only input to your results you actually own. Get the decisions right and the results follow, eventually, for exactly the reasons above.
That's the whole idea behind what I'm building at PokerLeap: train and check the decision, not the scoreboard. If you want to feel this variance in your bones, the outs and pot-odds calculators show you what a correct decision looks like on a single hand, which is the unit you can actually control, long before the results are allowed to tell you who you are.