One Qubit, Two Verdicts: Pure Dephasing With No Classical Rate Density, Energy Relaxation With One — Complete Monotonicity Measured Without Fitting on Superconducting Hardware; Two Predictions Refutable Against Existing Literature; and a Self-Audit of the Framework That Produced Them
The criterion, measured. This version adds the experiment the programme lacked: Pollard's criterion applied directly to a real decay curve, without fitting, and it withdraws a way of reading every fitted exponent in the corpus — including our own. 72 pages, 27 sections in 7 parts, 8 appendices. Eloy René Becerra Daly What is measured 20 non-adjacent qubits of one superconducting device — one session, one fixed layout, 39 circuits, 958,452 shots. The criterion answers differently in two channels of the same qubit. Energy relaxation admits a positive rate density (non-negative least squares over 200 free rates, χ²/dof = 0.99, 19/20 replicates compatible); pure dephasing does not (χ²/dof = 12.43, 17/20 refuted). The cause needs no model: the echo curve rises, in 10 of 220 consecutive segments above 3σ and six above 5σ, peaking at z = +11.01, across 7 of 20 qubits, while the T1 control gives 0 of 220 and backflow exactly zero in all twenty. Three independent confirmations: the Breuer–Laine–Piilo witness on the same counts, the residual against a one-rate Lindblad twin (+10.6σ), and frev = 0.674 failing to correlate with that witness (ρ = −0.107). What is withdrawn Reading a fitted β without reporting its goodness of fit. At σ/C ≈ 0.011 the two-parameter Kohlrausch form is rejected in both dephasing arms (χ²/dof = 11.49 and 1598), so its exponent licenses nothing in either direction. A non-rejection at low statistical power is not evidence of adequacy. Every β-dependent quantity of the new run. The pre-registered positive control fired: the T1 arm returned β = 0.8664 ± 0.0285 where a single rate must give 1. Its premise proved false — T1 is not a single rate on real hardware — but a premise falsified after the fact does not annul the clause. The distance to the pole and the branch assignment are withdrawn for that run. The universal claim that no platform violates β < 1, which survived into the conclusion of v2.0.0 after being corrected elsewhere. What is falsified in the previous version §20 of v2.0.0 states that revivals are slowdowns and not reversals, and that no local operation can revert the decoherence arrow. Both are falsified by measurement: the echo rises with the control at zero, and a single X pulse — a local unitary — reverts 67 % of the free-induction dephasing. The backflow bound of §3, which admits recoherence and bounds it, is the section the data support; it receives its first experimental test and is not refuted. What is unchanged The exact identities, the no-go theorem, the transferable criteria and the dark-matter conjecture stand as in v2.0.0, with that version's errata intact. The decisive test the programme identifies — β = (1 + αdiff)/2 on a single sample, refutable against existing NMR and diffusion literature without new experiments — is still not performed here. Scope, stated because it is easy to overstate One device, one session, twenty qubits, and a single-qubit observable blind to entanglement by construction. The measurement establishes that the criterion is applicable to hardware and that it separates two channels of the same qubit. It establishes nothing about superconducting qubits in general, and the corrected pre-registration for a second device is published rather than attempted. Reproducibility Every number of the new section is regenerated from the raw counts by a single extractor that writes them with the SHA-256 of the counts file, and a verification mode fails if any has moved. The figures are produced by a script that refuses to write if its own numbers disagree with that file. The register documents 44 inconsistencies, of which 20 remain open, each with the observation that would close it; one was closed by returning to hardware and one was falsified there.