Guide
Re-run our numbers for $0.
A job ID proves a run happened; it does not prove the number. The real way to check us is to run the experiment yourself — the flagship circuit is depth 7 and costs nothing.
What you need
A free IBM Quantum account on the open plan, which includes enough monthly runtime for everything below. Our flagship circuit is depth 7 with two entangling gates — among the cheapest experiments it is possible to run on real quantum hardware.
The measurement, step by step
1. Prepare the encoded Bell state. Encode a two-qubit Bell state into the [[4,2,2]] error-detecting code on four physical qubits of a Heron-class processor — pick a well-calibrated native coupler pair from the day’s calibration data. The [[4,2,2]] construction is textbook (Gottesman’s four-qubit code) and published prior work on several platforms; we claim nothing novel about it.
2. Measure three logical bases. Run three circuits — one per logical correlator (ZZ, XX, YY) — at 20,000 shots each, applying the appropriate basis rotation before measurement.
3. Post-select. Keep only the shots whose two pair-parities agree — the error-detection check of the code. Record the acceptance rate; ours was ~96% per basis. Report it with your number, always.
4. Compute the fidelity. F = (1 + 〈XX〉 + 〈ZZ〉 − 〈YY〉)/4 from the post-selected correlators. For a perfect Bell state, 〈XX〉 = 〈ZZ〉 = +1 and 〈YY〉 = −1.
5. Read your result honestly. On a good coupler pair you should land near 99.9% — and the number you get characterizes the pair you got, which is exactly the caveat we attach to our own: this measurement is functionally the two-qubit Bell fidelity of one native coupler, seen through the code’s coincidence filter. If your number differs from ours, that is the experiment working — different pairs, different days, different calibrations.
Want our exact artifacts?
We will send the reproduction package — the exact circuits as run, the pre-registered decode conventions, and the raw per-shot arrays behind the published confidence interval — to anyone who asks: team@atomiccubit.com, subject Reproduction package. Our published number re-derives offline from those arrays.
And if you run it and believe we got something wrong — say so. We publish corrections with dates on the ledger; it has happened before and it will happen again.