We Simulated 44 Million Casino Bets — The Real RTP of Plinko, Baccarat, Wheel and Blackjack

Casino games publish their RTP, and the maths behind them is well documented — but published claims deserve independent verification. So we ran the numbers ourselves: 9 million Plinko drops across every risk and row configuration, 10 million baccarat hands with exact Punto Banco drawing rules, 15 million Wheel spins, and 10 million blackjack hands comparing perfect basic strategy against the ways people actually play. Here is everything we measured, with a methodology anyone can reproduce.

Headline results

• Plinko returned 97.99%–99.03% in every configuration — the ~1% house edge of Stake-style tables is real, consistent, and independent of your risk/row choice.
• The 1000x slot hit 31 times in a million drops — matching the theoretical 1 in 65,536 per side almost exactly.
• Baccarat’s banker bet measured a 1.04% house edge over 10 million hands — the published ~1.06% figure verified within sampling noise.
• Wheel returned 98.53%–99.53% across all 15 risk × segment tables (15 million spins) — and the high-risk jackpot hit at 99.5%–100.5% of its theoretical frequency.
• Blackjack basic strategy measured a 0.44% house edge over 8 million hands — the lowest in the study — while “mimic the dealer” and “never bust” measured 5.68% and 5.94%: playing by feel costs about 13 times more.

Plinko: 1 million drops per configuration

We simulated 1,000,000 drops for each of the nine risk × row settings, using the same payout tables as our free Plinko table (matched to Stake’s published values). Each drop is a sequence of independent left/right bounces — the binomial model. Results:

Risk	Rows	Analytic RTP	Measured RTP (1M drops)	Max multiplier	Max-slot hits
Low	8	98.98%	98.94%	5.6×	7,759
Low	12	98.98%	98.95%	10×	476
Low	16	99.00%	98.98%	16×	36
Medium	8	98.91%	98.87%	13×	7,811
Medium	12	98.99%	99.00%	33×	514
Medium	16	98.99%	99.03%	110×	27
High	8	99.06%	98.90%	29×	7,725
High	12	99.12%	97.99%	170×	422
High	16	98.98%	98.61%	1000×	31

Three things stand out:

1. Every measured RTP sits close to its analytic value. The visible wobble on high-risk settings (97.99% on high/12) is exactly what extreme-variance configurations do to a million-drop sample: a handful of 170× hits more or fewer moves the total. That is the lesson of high risk.
2. Edge-slot frequencies match theory. On 8 rows the two outer slots (1/256 each) hit 7,725–7,811 times per million — theory says 7,813. On 16 rows: 27–36 hits against a theoretical 30.5. The board is exactly as fair, and exactly as brutal, as the binomial maths says.
3. Risk and rows do not change your expected return — only the shape of the ride, as our Plinko odds guide explains in detail.

To put the 1000× in human terms: at one drop per second, non-stop, you would wait about nine hours on average between 1000× hits — and lose steadily on the centre slots while waiting.

Baccarat: 10 million hands, exact rules

We dealt 10,000,000 hands from a continuously reshuffled eight-deck shoe, applying the full Punto Banco third-card table — the same rules as our free baccarat table. Results:

Outcome	Measured frequency	Theoretical
Banker win	45.867%	45.86%
Player win	44.610%	44.62%
Tie	9.523%	9.52%

Implied house edges from the measured frequencies:

Bet	Measured edge	Published
Banker (after 5% commission)	1.04%	~1.06%
Player	1.26%	~1.24%
Tie (8:1)	14.29%	~14.36%

Every figure lands within normal sampling variance of the published values. The banker bet’s superiority is not folklore — it falls straight out of ten million dealt hands, which is why “bet banker” is the entire content of honest baccarat strategy.

Wheel: 1 million spins per configuration

After shipping our free Wheel game, we ran its exact payout tables through the same treatment: 1,000,000 spins for each of the fifteen risk × segment configurations (15 million total). Every table is built to sum to 0.99 × segments — an analytic RTP of exactly 99.00% — so this is a clean test of whether a uniform random index behaves.

Risk	Segments	Analytic RTP	Measured RTP (1M spins)	Jackpot hits (measured / expected)
Low	10–50	99.00%	98.98%–99.08%	—
Medium	10–50	99.00%	98.85%–99.10%	—
High	10	99.00%	98.66%	99,660 / 100,000
High	20	99.00%	99.00%	49,998 / 50,000
High	30	99.00%	98.53%	33,174 / 33,333
High	40	99.00%	99.53%	25,134 / 25,000
High	50	99.00%	98.75%	19,950 / 20,000

Two observations:

1. Low and medium risk hug the analytic value (±0.15 points) because their variance is small. High risk wobbles more (98.53%–99.53%) for the same reason high-risk Plinko does: when the entire return rides on a 1-in-N jackpot, a few hundred hits more or fewer move the total visibly — across a million spins.
2. Jackpot frequencies are exactly uniform. Every high-risk jackpot count landed within 0.5% of its theoretical 1/N rate. There is no “due” segment and no hot streak — just the flat distribution our Wheel odds guide describes.

Blackjack: 10 million hands, strategy vs gut feel

Added July 2026. Blackjack is the one game in this study where your decisions change the house edge, so we tested that directly. Rules, stated in full: eight decks, dealer stands on soft 17, blackjack pays 3:2, double on any two cards, double after split allowed, resplit up to four hands (split aces receive one card each), no surrender, dealer peeks for blackjack. We ran 8 million hands of perfect basic strategy and, for comparison, 1 million hands each of two gut-feel strategies people actually use — same rules, same flat bet.

Strategy	Hands	Measured house edge (per initial bet)	Expected loss per $1,000 bet
Basic strategy	8,000,000	0.44%	$4.36
”Mimic the dealer” (hit to 17, never double/split)	1,000,000	5.68%	$56.78
”Never bust” (stand on hard 12+, never double/split)	1,000,000	5.94%	$59.35

Three findings:

1. Basic strategy’s 0.44% is the lowest house edge in this entire study — lower than baccarat’s banker bet (1.04%) and the ~1% of Plinko and Wheel. The analytic value for this rule set is roughly 0.43–0.45%, and 8 million hands landed right on it. Counting the extra money placed on doubles (10.4% of hands) and splits (2.8%), the cost per unit actually wagered was 0.39%.
2. Playing by feel costs about 13 times more. Both naive strategies gave back more than 5.6% of every initial bet. The gap — roughly five percentage points — is the measured value of basic strategy: it comes from doubling and splitting at the right moments and from hitting stiff hands against strong dealer cards, not from any single dramatic play.
3. “Never bust” measured worse than blindly mimicking the dealer (5.94% vs 5.68%). Refusing to bust feels safe, but standing on 12–16 against a dealer 7-through-ace simply hands those hands to the house. Busting is not the thing to avoid; misplaying is.

Two calibration checks confirm the engine: the dealer drew a natural in 4.743% of hands against a theoretical 4.745% for eight decks, and per resolved hand basic strategy won 43.5%, lost 47.9% and pushed 8.6% — in line with published blackjack distributions.

Methodology (reproduce it yourself)

• Environment: Python 3, standard library only (random, math).
• Seed: random.seed(20260703) — fixed, so every number above is exactly reproducible.
• Plinko: each drop = rows independent fair coin flips; slot index = number of rights; payout from the exact Stake-style tables used on our play page. 1,000,000 drops per configuration; analytic RTP computed as Σ C(rows,k)/2^rows × payout(k).
• Baccarat: eight-deck shoe, reshuffled below 20 cards; exact Punto Banco player and banker third-card rules; naturals end the hand; 10,000,000 hands.
• Wheel: uniform random segment index; payout tables identical to our play page (low/medium repeat a 10-segment block summing 9.9; high pays a single 0.99 × N jackpot). 1,000,000 spins per configuration, seed 20260703 (NumPy default_rng).
• Blackjack: eight-deck shoe, reshuffled below 52 cards; dealer stands on soft 17; blackjack pays 3:2; double any two cards, double after split, resplit to four hands, split aces one card each; no surrender; dealer peek. Total-dependent basic strategy for this rule set; flat 1-unit bets. 8,000,000 basic-strategy hands plus 1,000,000 each for the two naive strategies as defined above; seed 20260716 (added in the July 2026 update).
• No cherry-picking: these were single runs at the stated seed, reported in full.

What this means for players

The house edge is small, real, and unbeatable by settings or systems — but it is also honest: the games behave exactly as their published maths says. We keep a ranked table of every major game’s edge — measured where possible — in our casino house edge statistics page. Verify claims where you can (our provably fair guide shows how to check individual results at crypto casinos), treat big multipliers as the long shots they measurably are, and play only what you can afford to lose, where it is legal for you.