If you're practicing dice setting, you need a way to know whether your throw is actually different from random — and that means tracking data. Gut feeling is useless here. A few good sessions feel meaningful in the moment and mean almost nothing statistically. The SRR, or sevens-to-rolls ratio, is the standard metric the dice-setting community uses to measure whether a shooter is producing fewer sevens than chance would predict. This page explains what it is, how to calculate it, and — the part most guides skip — how much data you actually need before that number tells you anything at all.
What the SRR Is
SRR stands for sevens-to-rolls ratio. The formula is simple: divide your total number of rolls by the number of sevens you rolled. A purely random shooter should land a seven once every six rolls, producing an SRR of 6.0. That's because there are six ways to roll a 7 out of 36 possible two-dice combinations — a probability of exactly 1 in 6.
An SRR above 6.0 means you're rolling fewer sevens than random expectation. An SRR below 6.0 means you're rolling more. Practitioners typically treat a sustained SRR of 6.3 or 6.5 as the first threshold worth paying attention to — something meaningfully above the random baseline. The word "sustained" is doing a lot of work in that sentence, and it matters more than the number itself.
How to Calculate Your SRR
The formula is: SRR = Total Rolls ÷ Sevens Rolled.
Example: if you threw 300 times in a session and rolled 44 sevens, your SRR is 300 ÷ 44 = 6.82. If you threw 300 times and rolled 52 sevens, your SRR is 300 ÷ 52 = 5.77 — worse than random.
The simplest way to track this is tally marks in a small notebook. One column for total rolls, one for sevens. You don't need anything more complicated than that. What matters is that you write it down during or immediately after the session rather than reconstructing it from memory — memory favors the good sessions.
Keep practice sessions and casino sessions in separate logs. Home conditions differ from a live table in ways that matter: the backstop, the felt, the surface consistency, the fact that you're not rushing between shooters. Numbers you build at home may not transfer to the casino, and mixing the two datasets obscures both. Also note the dice set you used on each session, and any significant change in your throw style or conditions. If you switch sets mid-experiment, your data is measuring two different things.
The Sample Size Problem
This is where most beginners go badly wrong, and it's the most important section on this page.
Small samples are meaningless. Not "less meaningful" — meaningless. With only 100 rolls, a completely random shooter can easily produce an SRR of 7.0 or higher through pure chance, and there's no way to distinguish that from skill. Think of it the way you'd think about flipping a coin: if you flip it 10 times and get 7 heads, you don't conclude the coin is biased. You know the sample is too small to say anything. Ten heads out of 10 flips would get your attention. Seven out of 10 tells you nothing.
The Wong test, which is the most cited real-world controlled shooting experiment, used 500 rolls as a starting point — and even that was widely considered too small by statisticians to draw firm conclusions either way. The 2020 machine study, which used a purpose-built mechanical thrower designed to replicate controlled shooting with more precision than any human hand, ran 7,557 throws before testing for significance — and still didn't produce a statistically significant deviation from randomness. Seven thousand throws from a machine, and the result was inconclusive. That context should inform how you read your own 200-throw session.
Here's a practical reference for interpreting your sample size:
| Sample Size | What It Can Tell You |
|---|---|
| Under 500 rolls | Nothing meaningful — random variance is too wide |
| 500–1,000 rolls | Very preliminary — interesting but not conclusive |
| 1,000–3,000 rolls | Starting to matter — trends worth watching |
| 3,000–10,000 rolls | Meaningful data — patterns here are harder to explain as pure chance |
| 10,000+ rolls | Strong evidence either way |
Don't draw conclusions from fewer than 1,000 rolls. The range of 3,000 to 5,000 is where your cumulative SRR starts to be worth taking seriously. At 10,000 throws, you have data that's genuinely hard to wave away as variance. Most casual practitioners never get there — which is itself a reason to be cautious about what you conclude from your numbers.
Understanding Variance — Why Your SRR Will Jump Around
Even a completely random shooter's SRR bounces around from session to session. That's not a sign something unusual is happening — it's exactly what randomness looks like at small scales. One good session at SRR 7.2 and one bad one at SRR 5.8 average to roughly 6.5. Whether 6.5 over those two sessions is signal or noise depends entirely on how many total rolls are behind it.
This is why evaluating by session is the wrong approach. A single session is too noisy to tell you anything. What you want to watch is your cumulative SRR — the total-rolls-divided-by-total-sevens number that includes every throw you've ever logged, not just last Tuesday's. If you have a spreadsheet or a log, a chart of your running cumulative SRR over time is more informative than any individual session number. A genuine sustained edge should show up as the cumulative line staying above 6.0 consistently as the sample grows. Random variance tends to converge toward 6.0 as the sample grows. That's the difference you're looking for, and you usually can't see it until you have a few thousand throws in the log.
What to Log Beyond SRR
Total rolls and sevens are required for the SRR calculation — everything else is context that makes the number more interpretable over time.
Log which dice set you used. If you're testing the hardways set versus a 3V set versus a come-out set, they need separate logs — the whole point of sets is that they're designed to affect different numbers differently, and mixing their results defeats the purpose.
Separate practice and casino entries as noted above. Beyond that, note whether a given throw landed on-axis — if you can tell when you've drifted off-axis, an argument can be made that off-axis throws should be excluded from controlled-shooter data entirely, since an off-axis throw is by definition not executing the technique. Some practitioners log on-axis and off-axis throws separately and calculate SRR for each group. That's a more rigorous approach if you have the discipline to track it.
Table conditions are worth noting when you can: felt type, backstop condition (new pyramid rubber versus worn smooth), table length. These affect the bounce, and a consistent throw on a worn backstop may produce different results than the same throw on a fresh one. Your physical state matters too — tired, rushed, and distracted throws are genuinely less consistent, which contaminates your data if you lump them in with focused, deliberate practice sessions.
Using the Simulator as a Baseline
The simulator on this site generates purely random rolls. That makes it useful as a baseline for understanding what random variance actually looks like at whatever sample size you're working with.
Run 1,000 or more simulated rolls and track the SRR the way you would in a real session. Do it a few times. You'll see that the SRR wanders — sometimes it'll land around 6.2 for a run, sometimes around 5.8, occasionally higher or lower. That range of natural variation is the noise your real-world SRR has to consistently beat before it's evidence of anything. The exercise is useful because it makes the abstract statistics concrete: you can see for yourself how much a random SRR can look like a good SRR over a short sample, and how much sample you'd need before the difference becomes unambiguous.
Being Honest With Your Data
Confirmation bias is real, and it's a particular problem for self-testers. People tend to remember the sessions where the numbers were encouraging and underweight the sessions where they went the other way. If you only log when you feel like you threw well, your data is worthless — you've preselected for the good samples.
Log everything. Every session, including the ones where your SRR came in at 5.7. Especially those. The credibility of any self-reported dice control data depends on whether the person reporting it tracked the unflattering sessions with the same discipline as the good ones. That's harder than it sounds.
If your cumulative SRR after 2,000 throws is 6.1, that's not meaningful evidence of control. It's a number that falls within what a random shooter would produce. If your cumulative SRR is 6.8 after 5,000 throws, that's a different conversation — not proof, but the kind of data that's worth continuing to build on and worth looking at critically. The honest standard isn't "my SRR is above 6.0" — it's "my SRR is above 6.0 over a sample large enough that random variance can't explain it, and I logged every session, good and bad, to get there."