← What’s Lost with a Look
Pair run with raw readout, 21 October 2026
Does the entanglement cost of untouched Bell pairs survive (1) with every measured neighbour verified in its 0 cluster from the raw readout signal, and (2) against extra elapsed time, by comparing 39 measured qubits with 1 on the same schedule? Plan, circuits and analysis committed and pushed before submission.
plan · run_pairs_iq.py (uses run.py) · analyze_pairs_iq.py · SHA-256
Results, read by the committed analysis (full output)
Order of events, from the servers’ own records: GitHub received the plan, circuits and analysis at 01:15:30 UTC (22 Oct); IBM created the six jobs from 01:15:49 to 01:29:58 (kingston dbcm7hfi0auc73dv8680, dbcm8oni0auc73dv87q0; marrakesh dbcmae8c77vs7386ihu0, dbcmbl0c77vs7386ikg0; fez dbcmcv9gv2cc73eal2og, dbcme5hgv2cc73eal4cg), charging 35–37 seconds each.
- ibm_marrakesh: after 4 readouts, Bell pairs keep 0.0093 less concurrence than after the same idle in shots where every measured neighbour was verified in its 0 cluster on every readout, and the loss with 39 qubits read out exceeds the loss with 1 by 0.0088 on the same schedule (both copies agree): shots with a neighbour read as 1 are removed and extra elapsed time is excluded (a neighbour leaked to a level that reads as 0 is not; see the flip checks).
- ibm_kingston: clean-shot loss +0.0100 after 4 readouts (shots with a neighbour read as 1 removed); 39 read out exceeds 1 read out by 0.0104 (extra time excluded). The clean-shot reading cannot tell on kingston (copies disagree); the all-shot reading is above zero.
- ibm_fez: clean-shot loss +0.0070 after 4 readouts (shots with a neighbour read as 1 removed); 39 read out exceeds 1 read out by 0.0065 (copies disagree by the committed rule, so cannot tell with the pooled value shown; the repeat, +0.0068 [−0.0000, +0.0090], is consistent with zero).
- Where the committed analysis printed “copies disagree”, the copy-to-copy difference contains 0 in every case but one (fez, 2 readouts, not used above): one copy is wider, not shifted. rule · script · output
- Shots kept by the selection 97–99 %; mid readouts outside both clusters 0.2 % (fez), 0.4 % (kingston), 1.5 % (marrakesh).
Job records: kingston 0 · 1 · marrakesh 0 · 1 · fez 0 · 1; IBM calibration at firing: kingston · marrakesh · fez. Raw IQ files on request.