Answers are grounded in this paper and cite their sources.
Papers 7 and 8 of this series ran user-written circuits on IBM Heron devices through an operating layer (psf-os) that places a circuit from the coupling graph and the failed elements only, adds its own controls to each job and applies no error mitigation; a result reads when its mean squared error is at most half the mean square of the exact answer. This paper asks what decides whether a raw result reads. We split the error exactly as MSE / mean square = (1 - r)^2 + rest, with r = sum(z e) / sum(e^2) the share of the exact pattern the device keeps. Since the rest is never negative, r < 1 - 1/sqrt(2) = 0.293 means "does not read" by construction; the measured statement is the converse. Across every result of the series with an exact answer (179), including four of IBM's 2023 utility observables run raw (reading to 8 steps for <Z_106>) and the 1D Fermi-Hubbard model at U = 4 (reading at every size from 40 to 120 qubits at 4 Trotter steps, and at neither of the two sizes run at 6), 160 of the 166 results with r >= 0.293 read; the six that did not have a wrong shape as large as the shrink, or sit at the line. So whether a result reads is mostly whether r stays above 0.293, and the question becomes what sets r. A declared test on two devices found that it is not the CZ depth alone: at equal depth an 81-qubit XX chain kept less than a 33-qubit one, by up to 0.18. Where the 33-qubit chain is the first 33 qubits of the 81-qubit one, the extra loss on the same qubits opens next to where the longer chain continues and moves inward with depth; a repeat in a new job gave the same order, while a simulation with uniform local noise gives no gap. Finally, we state a problem rather than a method: dividing raw values by a measured share (of a full-signal circuit, or of a Clifford twin with the same pulses) cannot undo what is not a shrink, and on our records it changed few verdicts and made some results worse. No quantum advantage is claimed: every answer is computed exactly on a classical machine.
No takes yet. Share an insight, caveat, or question.
TN.Holdings (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: