Why they redesigned the circuit

The preprint submitted to arXiv on 28 July 2026 sets out to combine the complexity-theoretic hardness of sampling-based proposals with two capabilities: suppressing hardware errors and verifying the quantum computation itself. The authors' answer is structured circuits, which alongside provable hardness guarantees admit an encoding in a quantum code. That makes it possible at once to reach high fidelities at high circuit depths and to certify an experimental fidelity by measuring code syndromes.[1]

Set the parts out one by one. The demonstration runs on a 70-qubit, depth-70 Clifford circuit doped with 468 T gates. A total of 97 physical qubits encode that computation in spacetime codes, and gate error rates are suppressed by 10 times after syndrome post-selection. The 70 therefore gives the width of an encoded computation more than a hardware count; the real number on the hardware side is 97.[1]

The price of the certificate

The figure given for the result is a lower bound: 0.284 at 95 percent confidence for the fidelity of the resulting state. It says the fidelity is at least that much and leaves open how much higher. The authors also state that the certificate is device dependent. What is gained in return is concrete: the certificate requires substantially weaker noise assumptions than existing fidelity proxy benchmarks.[1]

The cost of the discard step sits here too. Syndrome post-selection means throwing away runs whose syndromes do not come back clean; the 10-fold suppression is bought with those discards, and the preprint does not hide the cost. On the comparison side stand the 2,415 logical two-qubit operations and roughly 15 minutes that Phys.org reports. On the classical side a prohibitive runtime is a moving threshold: a better simulation algorithm pulls it down, and the work has not yet been through peer review.[1]

The distance to a useful qubit

What this circuit produces is samples from a hard distribution; in the authors' own description the method is a systematic way of promoting a stabilizer state to a magic state while keeping an error-detected fidelity certificate. For that to become an application, the same encoding and discard scheme would have to be shown running on a computation whose output is valuable in itself, at a tolerable discard rate. Two signals are worth watching: whether an independent classical simulation attempt catches this run, and whether the preprint becomes a peer-reviewed publication with the same numbers intact.[1]