Three-box weak values, 22 October 2026

In the three-box weak measurement the box with the negative weak value reads about half of its textbook value (−0.53 on ibm_fez over a fourfold range of probe strength, 11 October; −0.50 to −0.69 on three processors, 24 September), while the positive boxes read about 0.8. This run asks whether that shortfall follows the negative sign or the |11⟩ corner, by moving the negative box (variants V11, V10, VA), and compares every box with a standard noise model (qiskit-aer) built from IBM’s calibration in force when the job fires. Plan, circuits, analysis, known-good check and synthetic test were pushed before any job.

plan · tb_common.py · run_threebox.py · analyze_threebox.py · synth_threebox.py · synthetic results · known_good.py · known-good check · hashes: plan, code

Fired 22 October: ibm_fez (qubits 142, 143, 144; 7000 shots per circuit) and ibm_marrakesh (6, 7, 8; 6000 shots): fez job, marrakesh job; noise-model values from the calibration at firing, saved before any result was read: fez, marrakesh.

Addendum, written after the jobs were submitted and before any result was read: the question has four answers (sign, corner, both, neither), not two; and an outside referee of the 11 October data found that a standard noise model plus one common probe offset reproduces those readings, so the offset is now a reading of its own. addendum · addendum_fourstate.py · planted test · referee report · hashes

Results (ibm_marrakesh; the fez job was cancelled while queued, why)

Probe offset. In the box whose ideal probe value is 0, the probe reads +0.049 to +0.081 (±0.010) in the three variants; the standard noise model built from IBM’s calibration at firing gives about +0.005, and IBM’s stated readout asymmetry of the probe qubit (q8) adds +0.006. The same offset appeared on ibm_fez on 11 October (+0.062) and in the 24 September runs (+0.060).
The negative box reads −0.59 to −0.60 against the model’s −0.76; the positive boxes also read 0.06 to 0.18 below the model. The shortfall relative to the model is shared by all boxes.
Sign or corner. The difference between the negative box and the positive boxes is consistent with zero in all three variants. The four-state reading names the sign on this one chip by a margin of 0.001; the addendum fixed before the data says a single-chip reading of that kind is not a finding. Not decided.

committed output · four-state and offset readings · outside referee · IBM calibration at firing

Bones · Wayfinder · Sage · Shuttle · Nova
Heuremen LLC
“We found it.”