Plinko guide
Crystal Plinko: what a seeded simulation shows
The crystal tag is just a page name, with no special board behind it. Here we use a computer simulation to watch a 13-row board in action, then hold the results up against the exact numbers.
How the simulation works
The simulation drops a ball through 13 rows. At each row a pseudo-random number decides left or right, with a 50/50 split. The slot is the count of right bounces. The slot's multiplier comes from a hypothetical pay table, and the net result of the drop is that multiplier minus the one-unit stake. A session is a sequence of drops; the simulation tracks the running total, its lowest point and the longest run of losing drops.
The pay table is invented for this page: 42.0x, 13.0x, 5.0x, 2.4x, 1.2x, 0.8x, 0.5x, 0.5x, 0.8x, 1.2x, 2.4x, 5.0x, 13.0x, 42.0x for slots 0 to 13. Its exact expected return is 98.46%, so the exact mean net per drop is -0.0154 units. It is not any operator's schedule.
| Session | Rounds won | Net (units) | Lowest point | Longest losing run |
|---|---|---|---|---|
| 1 | 26 | 4.70 | -7.30 | 25 |
| 2 | 42 | -6.00 | -9.80 | 12 |
| 3 | 44 | 15.50 | -9.70 | 11 |
| 4 | 35 | 17.80 | -5.30 | 12 |
| 5 | 48 | -0.60 | -1.60 | 11 |
| 6 | 43 | 11.50 | -9.80 | 9 |
Seeded pseudo-random simulation (Python random.Random, seed 4417), 6 sessions of 150 one-unit rounds. Average session result 7.15 units. This is a simulation, not a record of real play; a different seed gives different numbers.
Reading the simulated sessions
This is a simulation, not a record of anyone's play. Each session uses the same table and the same rules, yet the results differ a lot. Any drop that pays under 1x counts as a losing drop, and on this table that covers 73.32% of drops, which is why the losing runs get long. The exact expected net over 150 drops is -2.31 units; a session that lands far from it is not broken, it is normal.
Simulation against the exact answer
Simulations are useful for checking a calculation or exploring rules that are hard to solve by hand. They converge slowly, though. Running the same model with seed 4417 for 20,000 drops gives an average net of -0.0139 units per drop, against the exact -0.0154. With another seed, the figure would come out somewhat different. For plinko the exact answer is easy to get from the binomial formula, so the simulation is a check, not the source of truth.
Things a simulation cannot do
- It cannot tell you what the next real drop will be; it only shows the kinds of results a model can produce.
- It cannot reveal a hidden pattern in a fair game, because a fair game has none to find.
- It can mislead if it runs too few drops. A handful of sessions can sit well above or below the true average purely by chance.