Every poker player wants to know if they’re really winning or just running hot. The key stat most players watch is bb/100, short for “big blinds won per 100 hands.” It’s the standard way to measure win rate in poker. But short samples can be misleading—variance and luck can make you think you’re crushing when you’re only even. That’s why Bayesian statistics matter. This method updates your estimated win rate as you play more hands, showing how likely your current results reflect your true long-term edge.
What you actually estimate when you see bb per 100
Think of every block of 100 hands as one data point in a noisy process. Your win rate has an average value (the “mean”) and natural swings measured by standard deviation (SD)—a number that shows how much your results bounce above or below average. In Bayesian terms, you start with a prior belief—for example, assuming your true win rate is around 0 if you’re new—and then update that belief as you collect data. Each new sample narrows your uncertainty, producing a refined estimate of your true win rate that becomes more reliable over time.
In simple terms, a 95% credible interval means there’s a 95% chance your true win rate falls inside that range. It’s like saying, “Given what I know so far, I’m this confident about my edge.” That’s easier to grasp than the older “confidence interval” approach, which talks about imaginary repeated samples instead of the real data you have. This clarity is why Bayesian analysis is increasingly used by serious players who want honest, probability-based insight into their performance.
Why context from real money play matters for your posterior
Short samples behave very differently across real environments. A 6-max cash pool often shows SD near 80–100 bb per 100. Heads-up or very loose pools can be much higher, which slows convergence. Rake structures compress small edges, and table composition can introduce skew. Anonymous fast-fold pools change hand distribution and reduce table-selection effects. All of this makes context crucial when interpreting your Bayesian posterior.
Online poker formats differ wildly in pace and variance. On platforms offering real money poker, you’ll find options like anonymous tables, Quick Seat matchmaking, and fast-fold formats such as Zone Poker—each with its own rhythm and volatility. These differences directly affect your variance and how fast your statistical estimates stabilize. Playing real money poker and keeping track of your statistics as you go is key to understanding how you approach this game.
Your Bayesian update is only as good as the environment it models, and real money poker games are the only place where true variance and edge distribution appear.
Sharpening your fundamentals to reduce variance
You can often win more pots by learning about some of the techniques the pros use. In this video guide, you’ll see how small edges compound when fundamentals, position, and discipline align.
It explains why using range charts, mastering continuation bets, and knowing when to fold all reduce unnecessary variance. The cleaner your decisions, the more useful your Bayesian posterior becomes—because the model only works if the inputs (your plays) are consistent.
Worked example with realistic numbers and assumptions
Assume a prior mean of 0 bb per 100 and a prior standard deviation of 10. Your tracker shows a sample mean of 4 bb per 100 with an SD of 90.
After 5,000 hands, you have 50 blocks. The standard error (SE) of the sample mean is 90 ÷ √50 ≈ 12.7. The Bayesian posterior still overlaps zero widely. A 95% credible interval runs roughly from –20 to +28 bb per 100. Translation: you might still be break-even or modestly winning.
After 25,000 hands, you have 250 blocks. SE ≈ 5.7. The 95% credible interval tightens to about 4 ± 11 bb per 100. The data now outweigh the prior, and you can say with high confidence that you’re a small winner.
After 100,000 hands, you have 1,000 blocks. SE ≈ 2.85. The 95% credible interval is roughly 4 ± 5.6. At this scale, your posterior reflects a stable, repeatable edge rather than variance.
The math works out neatly in this case, which means you can track your updates in a simple spreadsheet without heavy coding.
How many hands for a trustworthy estimate
There’s no universal number for creating an estimate. It depends on SD and how precise you want your estimate to be. The common planning formula for margin around a mean is:
N = (1.96 × SD / desired margin)²
If SD = 90 and you want ±1 bb per 100 at 95% confidence, you need roughly 31,000 hands. In a Bayesian workflow, you don’t wait for a threshold—you update continuously and watch the credible interval narrow as you play more. The more hands you record, the sharper the posterior becomes.
| Parameter | Typical Range | Bayesian Effect |
| SD (bb/100) | 70–120 | Wider SD slows confidence gain |
| Hands played | 5k–100k | Each 10k narrows posterior width |
| Credible interval width | ~20 → 2 bb/100 | Shrinks with volume |
| Prior mean | –2 to +2 | Shapes early updates |
For live or mixed formats with lower hand counts, Bayesian updating is especially useful—it gives you a mathematically grounded sense of progress even before reaching massive sample sizes.
The real takeaway
In plain English, your win rate isn’t a fixed truth—it’s an evolving estimate that becomes clearer with every hand you play. Bayesian reasoning helps you update your confidence realistically, showing how likely your current results are to hold up in the long run. It’s not about removing luck; it’s about measuring how much of your outcome is luck versus skill. When you understand that balance, you gain the kind of confidence that lasts beyond the next session.
Mathematics can’t remove luck, but it can measure uncertainty. Once you quantify uncertainty honestly, your confidence becomes the kind you can trust—not the kind variance can break.