Does a mid-circuit readout advance an untouched neighbouring qubit’s phase more than an idle of exactly the same length? IBM 156-qubit processors, 21 October 2026.
Per chip: 39 measured qubits, 117 untouched qubits prepared on the equator. Each step is one of three modes — D: measured qubits idle for exactly the published readout length; M0: measured qubits read out, held in 0; M1: prepared in 1, then read out. Barriers on all qubits after every step so untouched qubits see identical scheduling in every mode. 0, 2, 4, 8, 16 steps; X and Y quadratures; 2000 shots; one job per chip. Ratio R0 = phase per step in M0 ÷ phase per step in D. Untouched qubits kept if their D slope is at least 1°/step (declared in advance; results without the cut are printed too).
The plan and analysis were written and hashed before any job was submitted (plan file 14:52:48, first job submitted 51 s later, local time): PREREG.md (served as .text so the bytes match the hash), SHA-256 0a8bc2b627a737f8… (full hash). Fez, Marrakesh and Kingston were the declared chips. Pittsburgh was added afterwards with the same hashed design and analysis; Boston was submitted and errored.
| chip | readout | R0 | 95 % interval | R1 − R0 |
|---|---|---|---|---|
| ibm_fez | 1.660 µs | 1.186 | 1.087–1.257 | +0.105 (above 0) |
| ibm_kingston | 2.180 µs | 1.204 | 1.106–1.343 | +0.153 (above 0) |
| ibm_marrakesh | 2.684 µs | 1.166 | 1.020–1.257 | −0.003 (consistent with 0) |
| ibm_pittsburgh (added) | 2.180 µs | 1.202 | 1.135–1.275 | +0.236 (above 0) |
Medians over untouched qubits, intervals from 2000 bootstrap resamples of untouched qubits within one job. Chips are the independent units.
The ratio above divides a phase excess by each untouched qubit’s own idle phase, so its value depends on which slowly precessing untouched qubits are kept. The excess itself is a phase per readout:
| chip | extra phase per readout, measured qubits in 0 (median, 95 %) | on untouched qubits precessing backwards |
|---|---|---|
| ibm_fez | +1.73° [1.29, 2.21] | +1.49° |
| ibm_kingston | +2.41° [1.96, 3.58] | +3.28° |
| ibm_marrakesh | +2.13° [1.44, 2.79] | +3.01° |
| ibm_pittsburgh | +1.79° [1.16, 2.66] | +2.75° |
Extra elapsed time would make the excess negative for untouched qubits precessing backwards; it is positive. The M1 arm here is compared with an idle whose measured qubits are in 0, so M1 − D includes a phase that appears without any readout when the measured qubits are in 1; the excitation test has the matching idle-in-1 arm and shows the kick is not larger with the measured qubits in 1 (kick in 1 minus kick in 0: +0.204 [−0.092, +0.320] on kingston, larger in 1 by at most 0.32°, consistent with zero; −0.673 [−0.982, −0.403] on marrakesh, smaller in 1). A regression of the excess on the idle phase (in kick_analysis.py) is biased by noise in the idle phase and is not interpreted. The planned reading 3 (constant ratio versus constant delay across readout lengths; plan hashed locally) cannot tell with three distinct lengths. kick analysis output · exploratory neighbour/far split
The plan’s SHA-256 was computed locally before submission and not externally timestamped. Job records: fez · kingston · marrakesh · pittsburgh · boston (submitted and errored)
| PREREG.md | design and analysis, hashed before submission |
| run.py | builds and submits the circuits (Qiskit 2.5.2, qiskit-ibm-runtime 0.49.0, SamplerV2) |
| analyze.py | the declared analysis; reads the raw files only |
| results_prereg.text | its output, with and without the cut |
| exploratory_neighbour_far.py | the neighbour / far split (exploratory) |
| raw_<chip>.npz | fez · kingston · marrakesh · pittsburgh — counts per circuit |
| roles_<chip>.json | which qubits were measured and which were untouched |
| job_<chip>.json | IBM job IDs and backend metadata |
| fire.log.text | submission log |
Bones · Wayfinder · Sage · Shuttle · Nova
Heuremen LLC
“We found it.”