Stake Originals Plinko: Probability Pyramid Reconstructed

Stake's Plinko has 16 rows. With a fair drop and an even peg deflection, ball paths follow a binomial distribution: C(16,k) divided by 2^16 for each landing bucket. The center bucket collects 12,870 of the 65,536 possible paths — 19.6 percent of every drop, by mathematical certainty, before any operator decides what to pay it. The published headline RTP for the game sits at 99 percent. We reconstructed the pyramid probability table from first principles and walked it back against the posted multipliers. The gap between what the binomial coefficient forces and what the marketing surface claims is the entire story of this game class.

That is the receipt. The rest of this note unpacks it.

What the Numbers Actually Say: The 16-Row Binomial Distribution

A 16-row Plinko board is, mathematically, a Galton box of depth 16. Every peg the ball strikes deflects it left or right with equal probability under a fair RNG implementation. The position the ball lands in is the count of right-deflections across 16 independent trials. That is the textbook definition of a binomial random variable with n equal to 16 and p equal to 0.5.

The number of distinct paths that lead to bucket k, counting from the far left at k equals zero, is the binomial coefficient C(16, k). The denominator is 2 to the sixteenth, which is 65,536. We reconstructed the full pyramid below. Numbers are raw path counts, then probability, then cumulative probability working in from the center.

Path counts sum to 65,536. Probabilities sum to one. The pyramid is symmetric — the board is a perfect mirror across bucket 8, and the math is forced. There is no operator discretion at this layer. Any provider's 16-row Plinko, audited by any of the labs published in the Gaming Laboratories International certificate library, must reproduce this distribution to pass an RNG statistical randomness test under the NIST 800-22 battery. That is the standard scope GLI publishes for game-math verification.

The two outermost buckets land 1 time in 65,536. That is roughly once every six and a half hours of continuous play at one drop per second. The reader's intuition about edge multipliers — that they hit "rarely" — under-counts how rare. The center bucket lands 128 times in the same span.

This is the floor. Everything an operator does sits on top of it.

What Nobody Mentions: Where Operator Margin Lives in the Bucket Layout

Stake's Plinko publishes three risk tiers — low, medium, high — and a configurable row count from 8 to 16. We focus on 16 rows because that is the configuration with the longest tail and the widest published multiplier spread. The marketing surface for the game class advertises an RTP at or near 99 percent across the tier menu. That number is the dot product of the binomial probability vector above and the multiplier vector the operator assigns to each bucket.

This is where the gap lives.

The multiplier vector is operator-set. The probability vector is mathematically fixed. So an operator that wants a 99 percent RTP, given the symmetric binomial weighting that concentrates 19.6 percent of drops in the center bucket and a further 17.5 percent in each of the two adjacent buckets, has very little choice about what the center bucket pays. Roughly 54.5 percent of all drops land in the three central buckets — 8, 9 and 7. The center multiplier on the high-risk 16-row table is published as 0.2x. The center on low-risk is 0.5x. Below one in both cases. Below one for 54.5 percent of all drops.

That is the operator margin engine. The hold is not collected at the edges. The edges are the spectacle. The hold is collected at the center, drop after drop, where the binomial forces the ball.

The published 1000x payout on the high-risk 16-row outer bucket, in isolation, sounds like generosity. Cross-reference it against the path count and you get its honest weight in the RTP equation: 1000 multiplied by 1 divided by 65,536, which contributes 0.01526 to expected return per drop. The center bucket at 0.2x contributes 0.2 multiplied by 12,870 divided by 65,536, which is 0.03927. The center, paying a fifth of stake, contributes more than two and a half times what the headline 1000x edge does to total expected value.

The standard marketing claim — that high-risk tiers offer "bigger wins" — is technically true and analytically empty. The mode of the distribution does not move when you change the tier label. Only the payout vector does. The probability vector is forced by the binomial coefficient, audited by the certification body, and identical across risk tiers. The published eCOGRA seal program scope covers game math verification against paytable specification — it confirms the multipliers match the published table. It does not, and cannot, change the underlying probabilities.

A small fieldnote. The Stake Plinko interface displays the multiplier value as the ball lands, with a brief flash on the bucket. The center buckets are dimmer in the default color palette. The far edges are saturated red and green. The visual weight is inverted from the probability weight.

The Real Cost: Per-1,000-Drop Reality Check Across the Three Risk Tiers

Take 1,000 drops at $1 stake. That is $1,000 wagered. The 99 percent RTP claim implies an expected return of $990. The expected loss is $10 — on average, over a long enough sample.

Translate that across the binomial distribution and the picture sharpens.

Of 1,000 drops, expect roughly 196 in the center bucket, 175 in each of the two adjacent buckets, 122 in each of the next pair out, and so on down the pyramid. The two outermost buckets — the ones the marketing screenshots feature — appear, on expectation, 0.015 times in 1,000 drops. Three plays out of every 200,000 land at the edge. Concession: when the edge does hit, it pays well. A high-risk 16-row edge return at 1000x on a $1 stake is $1,000, which is the entire 1,000-drop session in a single drop. That outcome is real. It is also, in the long run, what funds the marketing.

Now the teardown. The variance around the $10 expected loss is enormous. Across 1,000 drops at the high-risk tier, the standard deviation of return per drop, computed against the published multiplier vector, dwarfs the mean. A 1,000-drop session that ends down $400 is well within one standard deviation of expectation. A 1,000-drop session that ends up $300 is also within one standard deviation. The marketing-headline RTP is a long-run convergence value. It is not a session-level forecast. It is not a weekly forecast. For the typical player session at this risk tier, it is approximately useless as a guide to outcome.

Walk the cost out further. A player running 1,000 drops per day at $1 stake, every day, for one year — 365,000 drops, $365,000 wagered — expects to lose $3,650 to the 1 percent house margin at the published RTP. That is the floor. It assumes the operator's RTP claim is accurate to the third decimal, audited continuously, and that the player executes no behavioral mistakes (no chasing, no stake escalation, no tier switching mid-session). It also assumes the player can settle every transaction without payment friction — and the operator's segregated player fund holds up under the cumulative drawdown, which is a separate document and not one we have for Stake.

Add the real costs nobody counts. Time at $20/hr opportunity cost, 1,000 drops at one drop per 3 seconds is roughly 50 minutes, multiplied by 365 days, is 304 hours, valued at $6,080. The compounding effect of running deposits through card and crypto rails, which carry settlement spreads of 1 to 3 percent on a typical session deposit-and-withdraw cycle. The cognitive load of treating a binomial distribution as if it had memory, which it does not.

The honest number is not $3,650 lost per year. The honest number is closer to $10,000, once opportunity cost and rail friction are loaded in, and the headline RTP is treated as a long-run convergence value rather than a session-level promise. The published H2 Gambling Capital industry tracker put global iGaming GGR at $94bn for 2024 — that aggregate is, mechanically, the sum of every player's gap between expected return and actual outcome. Plinko sits inside that sum. The pyramid is one of the more elegant collection mechanisms in the catalog.

If You Only Remember One Thing

The probability vector is forced. The payout vector is chosen. Both are visible. Multiply them together row by row before you treat the headline RTP as a fact about your evening.

The center bucket of a 16-row Plinko collects 19.6 percent of every drop, by binomial certainty, before any operator decides what to pay it. That is the number. It is published in every elementary statistics textbook. It speaks for itself.

FAQ

How is the Stake Plinko probability distribution derived?

The 16-row Plinko board is a Galton box. Each peg deflects the ball left or right with equal probability under a fair RNG. The probability of landing in bucket k is C(16, k) divided by 2 to the sixteenth, which equals 65,536. The center bucket has C(16, 8) which is 12,870, giving 19.638 percent. The distribution is symmetric, and any audited implementation must reproduce it to pass NIST 800-22 statistical randomness tests in the standard certification scope.

Does the risk tier (low, medium, high) change the probability of where the ball lands?

No. The probability vector is a function of the binomial coefficient and is identical across all three risk tiers on the same row count. Only the multiplier assigned to each bucket changes. The high-risk tier widens the payout spread — smaller multipliers on the central buckets, larger on the edges — but the ball lands in the center 19.6 percent of the time regardless of which tier is selected. This is mathematically forced.

What does the 99 percent RTP figure actually represent?

It is the dot product of the binomial probability vector and the operator's chosen multiplier vector for that risk tier. It expresses long-run convergence: across many millions of drops, total return converges toward 99 percent of total stake. It is not a session-level forecast. A single session of 1,000 drops can end anywhere within a wide standard deviation band, and the 1 percent house margin only materializes over sample sizes far larger than any individual player accumulates.

How often does the 1000x edge multiplier on 16-row high-risk actually hit?

The outermost bucket on a 16-row board has exactly one path leading to it — C(16, 0) equals 1 — giving a landing probability of 1 in 65,536 per drop. At one drop per second of continuous play, that is once every roughly 18 hours. The 1000x contribution to total RTP is 0.01526. The center bucket's 0.2x multiplier contributes 0.03927, which is more than two and a half times the headline edge contribution.

Are these probabilities the same on every operator's Plinko implementation, not just Stake?

The binomial probabilities are mathematically identical across any 16-row Plinko with a fair RNG and symmetric peg deflection. What differs across operators is the multiplier table assigned to buckets, the row-count options offered, the published RTP target, and the certification body that signs off on the math. Cross-reference the certificate scope on the operator's published audit page before assuming two games are equivalent at the payout layer.

Why does the center bucket pay below 1x stake on every risk tier?

Because the center is where roughly 54.5 percent of all drops land — buckets 7, 8 and 9 combined. To target a 99 percent RTP given that concentration, the operator must pay below 1x on the central region. If the center paid 1x, the math forces total RTP above 100 percent, which no commercial operator will publish. The hold is collected in the center, drop after drop. The edges are the spectacle that funds player retention.

Can certification by GLI or eCOGRA change what these probabilities are?

No. Certification verifies that the RNG produces statistically random output and that the multiplier table matches what the operator publishes. It cannot alter the binomial coefficient. The published GLI audit scope covers RNG statistical randomness under NIST 800-22, game math verification against the paytable specification, and RTP empirical validation across a sample of simulated rounds. The probability vector is upstream of all three tests.

What is the most expensive misunderstanding players have about this game?

Treating the headline RTP as a session-level expectation. RTP is a long-run convergence value over sample sizes that no individual player ever reaches. The session variance, particularly on the high-risk tier, dwarfs the 1 percent house margin in any practical sample. A player running 1,000 drops per day for a year, fully loaded with opportunity cost and payment-rail friction, ends materially worse off than the headline RTP suggests — closer to a 2 to 3 percent effective drag once everything is counted.