The Receipt Machine
Ask a quantum computer for random bytes and it prints you a receipt: the Bell test that shows the device is quantum, the entropy it actually delivered, the price of what the readout threw away, and whether anything was left to certify. Sometimes the answer is nothing, and that's the machine being honest.
1 · Predict the bill
The engine only reports how often each outcome appeared, not in what order, so it charges log₂(shots!) bits for the lost order. Each shot brings qubits × h bits of entropy. Certify only what's left.
h is the conservative per-bit min-entropy (NIST SP 800-90B estimators, 99% confidence), taken from our real runs; a fresh run measures its own.
2 · Replay a real run
Three receipts printed on 27 September 2026, kept exactly as the engine returned them.
3 · Print a live receipt
Uses your Moth Atlas key through a small proxy (Moth's API only answers its own site). The proxy forwards three whitelisted calls and keeps nothing. qpu runs on real IBM hardware and costs credits (5 per run on 27 Sept).
Why a receipt?
Random bytes look the same whether they came from a coin, a computer or a qubit, so you can't tell by looking. You can test the machine instead. A Bell test puts two qubits in an entangled state and measures them at different angles. Anything classical scores at most 2; quantum mechanics allows up to 2√2 ≈ 2.83. Our IBM runs scored 2.61 and 2.70, about 26 and 30 standard deviations past the classical line.
Quantum still isn't the same as usable. Noise leaves each raw bit with less than a full bit of randomness, and the counts-only readout has to pay for the lost shot order. At the documented default of 12 qubits, our hardware run came up 601 bits short, and the engine printed zero bytes rather than pretend. At 64 qubits it certified 32 bytes, with its one untested assumption written on the receipt.