Plinko Odds: The Binomial Distribution Behind the Board

A plinko board is a Galton board. The ball takes n independent left-or-right steps, one per row, and the slot it lands in is the number of times it went right. That makes the landing slot binomially distributed: the probability of slot k on an n-row board is C(n,k) / 2ⁿ. On the standard 16-row board that means the centre slot takes 19.64% of drops and each outer slot takes 1 in 65,536.

Nothing about the animation matters. The bounces are decided before the ball is drawn, one bit per row, from the same hash chain that drives every other provably fair original.

Where the left-right decisions come from

Each row consumes one random number. The implementation pattern is the same one used to derive a dice roll from an HMAC digest: bytes from HMAC-SHA256(serverSeed, "clientSeed:nonce:cursor") are converted to floats on the unit interval, and each float is compared to 0.5 to give a direction. Sixteen rows, sixteen floats, and the slot index is the count of rights.

Because the source is a cryptographic hash over an incrementing nonce, the sixteen decisions are independent and each is a fair coin. That independence is what licenses the binomial formula, and it is the reason the seed and nonce mechanics are worth understanding before any of the probability makes sense.

The exact odds for every slot on a 16-row board

There are 2¹⁶ = 65,536 equally likely paths through the board. The number of paths ending in slot k is C(16,k), so the probability is that count divided by 65,536.

The three central slots absorb 54.55% of drops. The five central slots absorb 78.99%. The four outermost slots combined — the ones carrying every headline multiplier — take 0.052%.

Computing this yourself is a five-line job:

from math import comb

rows = 16
total = 2 ** rows

for k in range(rows + 1):
    p = comb(rows, k) / total
    print(f"slot {k:2d}  paths {comb(rows, k):5d}  {p * 100:8.4f}%")

Row count changes the sharpness, not the fairness

More rows means more slots and a narrower distribution, because the extremes require more consecutive identical decisions.

This is why a 16-row board can advertise a 1000× pocket and an 8-row board cannot. The multiplier has to be roughly the reciprocal of the probability for the sums to work, and 1 in 256 does not support a four-figure payout.

Why the huge edge multipliers barely matter

RTP is the probability-weighted sum of the multipliers:

RTP = Σ  p(k) × m(k)   where p(k) = C(n,k) / 2ⁿ

Take an illustrative high-risk 16-row table — constructed here to demonstrate the arithmetic, not copied from any operator:

1000, 130, 26, 9, 4, 2, 0.2, 0.2, 0.2, 0.2, 0.2, 2, 4, 9, 26, 130, 1000

Its slot-by-slot contribution to RTP, counting each pair once at double weight:

The 1000× pockets — the entire reason anyone selects high risk — contribute 0.0305 out of 0.9898, which is 3.08% of the game’s total return. Delete them from the table entirely and the RTP falls to about 95.9%. They are a rounding adjustment dressed as a jackpot.

Meanwhile the 5-and-11 pair, paying a modest 2×, contributes 0.2666 — nearly nine times as much. Return in plinko is generated in the middle of the board, always.

What risk levels actually do

Risk level does not touch a single probability. C(16,k) / 65536 is a property of the board geometry, not of a setting. What the risk selector changes is m(k) — the multiplier assigned to each fixed weight.

Three illustrative 16-row tables, each computed with the formula above:

Same probabilities, same return to within a tenth of a percentage point, radically different experience. Low risk gives you a slow bleed with frequent small returns; high risk gives you a fast bleed punctuated by occasional real money. This is the identical structure that appears in crash cash-out targets, where every target has the same expected value and only the variance moves, and it is the clearest practical illustration of why RTP and volatility are separate quantities.

The near-miss illusion

“It was one slot away from the 1000×” is the most common thing said about plinko and it is not a description of anything.

Slot 1 is hit 16 times more often than slot 0. Slot 2 is hit 120 times more often. Slot 3, 560 times. Landing in slot 1 is not a near miss on slot 0 in the way that missing a bus by ten seconds is a near miss — it is the overwhelmingly more likely of two outcomes that happen to sit next to each other in physical space.

The board’s geometry deliberately places the rarest outcome adjacent to a merely rare one, and the animation gives you a full second to watch a ball drift toward it. Both are design choices about presentation. Neither changes C(16,0) = 1.

Checking a real multiplier table

You can audit any published plinko table without trusting a single claim on the site:

1. Count the rows to get n, and count the slots to confirm there are n + 1.
2. Read the multipliers off the interface, left to right.
3. Compute Σ C(n,k)/2ⁿ × m(k).

If the result is 0.99, the game runs a 1% edge. If it is 0.96, it runs a 4% edge no matter what the help page says. If the table is asymmetric between left and right, that is worth a very hard look, because a symmetric binomial with an asymmetric payout table has no legitimate reason to exist.

That check tells you what the game costs. It does not tell you the drops are honest — for that you need to recompute the individual results, which is the verification procedure and is a separate exercise. And even both together leave the operator’s solvency untouched, which remains the limitation no cryptography addresses.