How Kaspa Blackjack Natural 3:2 Payouts and Fee Reserves Work

A Kaspa blackjack payout is settled by a covenant script on L1 — not by a server, not by a promise. When Arena Blackjack declares a natural 3:2 payout, that ratio is encoded in the Kaspa script that controls the game's UTXO before either player sits down. This article walks through how the covenant pre-locks stakes, liveness bonds, and a fee reserve so that every outcome — including the premium natural blackjack payout — is guaranteed by the chain itself.

Arena Blackjack is a public mainnet beta. The mechanics below describe how the program works, not an invitation to treat it as risk-free.

Why On-Chain Blackjack Needs Pre-Locked Payouts

In a traditional online casino, the house promises to pay 3:2 on a natural blackjack. You trust that promise. If the casino disappears, your balance disappears with it.

On Kaspa, there is no house balance to draw from. A game UTXO is a self-contained contract: everything the game can ever pay out must be locked inside it at the moment it's created. If the covenant doesn't pre-lock enough KAS to cover a natural blackjack payout, the game literally cannot pay it — the chain would reject the settlement transaction for violating conservation rules.

This is the fundamental constraint that shapes every number in Arena Blackjack. The covenant must lock enough to cover:

• The worst-case payout (a natural blackjack at 3:2)
• Liveness bonds (incentives for both sides to keep playing)
• Fee reserve (service fees, proof fees, network fees)

If any of these are underfunded, the game cannot settle. The covenant doesn't negotiate — it enforces.

Anatomy of a Game Lock: Stakes, Bonds, and Reserve

Every Arena Blackjack hand locks value in three distinct layers. Using a 5 KAS table as a concrete example:

Player side:

Dealer side:

Total game value: 22.5981 KAS

The asymmetry is deliberate. The player risks their stake plus a small bond. The dealer must lock significantly more because the covenant has no bankroll behind it — the UTXO *is* the bankroll. The 1.5× multiplier on the dealer's game stake exists precisely to cover the natural blackjack payout: if you bet 5 KAS and hit a natural, the covenant pays you 7.5 KAS in profit (3:2), and that money has to come from somewhere.

The real wager remains symmetric — 5 KAS versus 5 KAS. The extra dealer lock is not a larger bet; it's structural coverage the covenant requires.

The liveness bonds serve a different purpose entirely. They're anti-griefing deposits: if a player commits to a hand and then disappears — fails to reveal their seed or make a move within the deadline — they forfeit their bond. The PLAYER_SEED_TIMEOUT path is the only way a player loses their 2 KAS bond. It's the covenant's answer to "what if someone joins and never plays."

How the Covenant Encodes Natural 3:2

The game state is packed into a 240-byte ABI structure that the covenant script reads directly. This state includes the player's and dealer's hands, the game phase, the rules variant (S17 or H17), and the stake amount. The covenant doesn't interpret blackjack strategy — it checks terminal conditions against pre-committed branch amounts.

Each Room's covenant has terminal branches:

PLAYER_WIN   → player receives: stake + win (1:1, or 3:2 if natural)
DEALER_WIN   → dealer receives: stake + win
PUSH         → both receive: stake returned

And timeout branches enforced by DAA-relative timelocks:

PLAYER_SEED_TIMEOUT     → player bond forfeited
PLAYER_ACTION_TIMEOUT   → permissionless: anyone can claim
DEALER_TURN_TIMEOUT     → permissionless: anyone can claim

The natural blackjack check happens inside the terminal branch logic. A natural is defined as a two-card hand totaling 21 — an Ace plus a ten-value card (10, J, Q, or K). The covenant verifies this from the game state: if the hand is terminal, the total is 21, and exactly two cards were dealt, the payout branch uses the 3:2 multiplier instead of 1:1.

Crucially, a natural blackjack also beats a three-or-more-card 21. If both player and dealer reach 21, the one with fewer cards wins. The settlement display shows the card count and explicitly marks the natural — there's no ambiguity about which 21 wins.

The ZK proof (Groth16) serves a complementary but separate role. It proves the deck was shuffled fairly using a canonical Fisher–Yates shuffle over a BLAKE3-derived seed, each card was revealed from the correct Merkle position, and the root matches the committed state. But the ZK proof does not enforce payouts — that's the covenant's job. Fairness and rules are two distinct guarantees, verified by two different mechanisms.

The Fee Reserve: Keeping Settlement Possible

The fee reserve exists because on-chain games face a problem traditional casinos don't: variable network fees. A settlement transaction must pay Kaspa network fees to be included in the DAG, and those fees depend on congestion. If the game UTXO locked only stakes and bonds, a fee spike could make settlement impossible — the transaction would be too "heavy" to broadcast.

The reserve (3.0981 KAS on a 5 KAS table) covers three things:

1. Service fee — 0.9% of the two-stake pot, charged only to the winner. On a 5 KAS table, that's 0.09 KAS. 2. Proof fee — currently 0 for HOSTED mode (the operator runs the prover on their own hardware). The mechanism exists for future LOCAL prover fees. 3. Network fees — the actual cost of broadcasting the settlement transaction to the Kaspa DAG.

After settlement, any unused reserve flows back to the dealer. The reserve is a buffer, not a fee — it's returned if conditions allow. The settlement screen breaks this down explicitly: service fee, prover fee, network fee, and returned reserve are shown as separate line items.

Even on a push (tie), the service fee is deducted. The settlement display shows Push · stake returned alongside the exact fee amounts, so you can verify the conservation:

Game lock = player output + dealer output + service output
            + prover output + network fees

This conservation check is performed from the recorded settlement facts — the terminal transaction outputs are summed and compared against the opening game value. If the numbers don't add up, something is wrong. The verify drawer exposes this check for every completed hand.

How Kaspa Forge Arena Applies This

Arena Blackjack runs six OPEN tables at fixed stakes: 5, 10, 25, 50, 75, and 100 KAS. All current tables use S17 rules (dealer stands on soft 17) and the natural 3:2 payout. The lobby in Desk shows the rules and natural payout for each joinable room before you sit down.

When you join a table, the wallet interface shows your exact contribution (stake + bond) and breaks it down in a drawer: stake, bond, and proof fee as separate amounts. The dealer's lock is not part of your contribution — it's funded by the Room's covenant independently.

The settlement screen after a completed hand shows:

• Stakes in play — the symmetric wager (e.g., 10 KAS for a 5 KAS table)
• Contract locked — the full game value (e.g., 22.5981 KAS) broken into stakes, bonds, and reserve
• Settlement outputs — exact amounts to player, dealer, service, prover, and network
• Returned reserve — any unused portion flowing back to the dealer

For completed hands, a Verify drawer exposes the raw commitment, settlement txid, covenant branch, and the conservation check. The public post-settlement verifier can independently reproduce the deck proof from committed seeds — though as of v0.1, it verifies evidence consistency, not DAG inclusion or payout legality.

Honest Boundaries: What Beta Means

Arena Blackjack is a public mainnet beta. The covenant has passed five independent audits plus two P1.1 re-audits. The ZK proof has been verified on mainnet. Settlement transactions have been broadcast and confirmed. But the product is not risk-free, and the following boundaries matter:

• The player_hit_deal_timeout path has never been taken on mainnet. This is the permissionless timeout that pays the player if the dealer accepts a hit declaration and fails to deal the card. It exists in the covenant and has been tested in lifecycle probes, but no live hand has triggered it.
• Only Chromium-based browsers have been verified for WASM parity. Firefox, Safari, and mobile browsers have not been tested.
• The HOSTED proof fee is currently 0. The prover runs on the operator's hardware. This is a beta convenience, not a permanent guarantee — the fee reserve mechanism exists for future LOCAL prover fees.
• Network fee assumptions are untested at scale. The relay fee gate caps at 150 sompi/gram. Whether mainnet conditions sustainably stay below this threshold is not proven.
• The public verifier v0.1 checks evidence consistency, not chain inclusion. A green v0.1 verdict means the deck proof is internally consistent — it does not mean the settlement transaction was valid on-chain or that the payout amounts were legal.

These aren't disclaimers to skip over. They're the honest edges of what's been built and what hasn't been battle-tested yet. The covenant guarantees what it encodes — but "what it encodes" and "every possible real-world scenario" are not the same thing during a beta.