Games · Casino War
Casino War: the simplest edge in the house
The simplest game in the casino — one card each, higher wins — and it still tilts to the house. Deal a funded balance and you'll find the catch: on a tie you either surrender half your bet or double into a "war" that only pays even money on the raise. That asymmetry is the entire edge — an exact 2.88%. Play it, then watch that edge emerge over thousands of hands in the Lab.
Set your ante and deal.
It's a tie! Double your bet and go to war, or surrender half.
Free game, fun credits only — no real money. Optimal play: always go to war, never surrender, never take the Tie bet.
The dashed gold line is the exact house edge, 2.88%. Short runs swing wildly; the more hands you play, the tighter the observed edge hugs it — the house edge is a long-run certainty, not a per-hand one. For contrast: surrendering every tie is 3.70%, and the Tie side bet is a brutal 18.65%.
Casino War is the simplest game on the floor — one card each, higher wins. And yet the exact math, below, shows it still tilts to the house. Every figure is closed-form (no simulation) for a 6-deck shoe.
(23 in 311)
(same for the dealer)
What each bet really costs
| Bet / choice | House edge | Verdict |
|---|---|---|
| Go to war (on a tie) | 2.88% | Optimal play |
| Surrender (on a tie) | 3.70% | Costs 0.82% more — never do it |
| Tie side bet (10:1) | 18.65% | Sucker bet — ~6× the main edge |
Why it works out this way
- No tie (about 93% of hands): a coin flip — you win or lose one unit, and it washes out to zero on its own.
- A tie (7.40%): here's the catch. Go to war and you must double your bet, but a win only pays even money on the raise — the original ante just pushes. Win 54%‑ish of wars (ties go to you) for +1 unit; lose the rest for −2 units. That asymmetry is the entire house edge.
- Surrendering hands back half your ante every tie — a flat 3.70% drag, worse than the 2.88% from always warring. So the rule is simple: always go to war, never surrender.
- The Tie bet pays 10:1 but a tie hits only 7.40% of the time — a fair payout would be about 12.5:1. That gap is a brutal 18.65% edge. Skip it.
Does the deck count matter?
A little — more decks means a slightly higher tie chance, so a slightly higher edge. Still tiny either way:
| Decks | Go-to-war house edge |
|---|---|
| 1 | 2.42% |
| 2 | 2.70% |
| 6 | 2.88% |
| 8 | 2.90% |
How this simulation works
- Rules modeled
- Standard 6-deck Casino War. Player and dealer each get one card (ace high; suit is irrelevant to the game); the higher card wins even money. On a tie the player either surrenders (forfeits half the ante) or goes to war: raise an amount equal to the ante, the dealer burns 3 cards and deals one more to each — the player's card winning OR tying beats the dealer, paying even money on the raise while the original ante pushes (net +1 ante); a loss forfeits both (net −2 antes). Optional Tie side bet pays 10:1 when the first two cards tie. Stakes are fun credits in integer cents.
- Assumptions
- Hands are independent (a freshly shuffled shoe each round). Ace is always high; suits never break a tie — a tie is a tie. No side rules (no war-tie bonus). This models the common US 6-deck rules; other deck counts shift the edge slightly (shown in the Odds tab). Fun credits only, no real money.
- Mathematical basis
- Every figure is exact and closed-form, not simulated. P(tie) = (R−1)/(N−1) for R = 24 cards per rank and N = 312 cards; the go-to-war, surrender, and Tie-bet edges follow directly (2.88% / 3.70% / 18.65% at 6 decks — the standard published figures). The Play and Lab draw from the same engine; the Lab's Monte-Carlo only validates the exact edges, never redefines them.
- Engine version
- casino-war engine 1.0
- Validation
- 14 unit tests — exact edges vs the published figures (war 2.88% / surrender 3.70% / Tie bet 18.65% and P(tie)=23/311), the war payout mechanic (win +1 / lose −2 / ties to the player), surrender (−½), the 10:1 Tie bet, deck-count sensitivity, seeded determinism, and a 200,000-hand convergence to 2.88%.
- Last reviewed
- Verified by exact unit tests against the published house edges (the math is closed-form arithmetic, not a solver) — no separate independent-review gate was required for this engine. Last reviewed 2026-06-30.