GUESS / Symmetry-Decay Calibration Preregistration¶
Date: 2026-05-06
This preregistration prepares a hardware GUESS / symmetry-decay validation using parity leakage as a witness. It does not submit IBM jobs, reserve backend time, or authorise QPU spend.
Readiness Basis¶
The no-QPU readiness analysis is:
It fits log parity survival versus circuit depth as a cumulative-noise proxy. It is not zero-noise extrapolation, because the promoted Phase 2 datasets do not contain explicit folded-noise scale factors or another controlled noise knob.
Readiness signal:
- Phase 2 A+G even, odd, and sector-mean series have
R2 >= 0.9989. - Popcount-control state fits have
R2from about0.958to0.992. - These fits justify preparing a folded-noise protocol, not claiming GUESS mitigation from existing data.
Scientific Question¶
Can parity leakage serve as a stable hardware symmetry-decay witness for symmetry-guided zero-noise extrapolation on the same heterogeneous Kuramoto-XY family?
Claim Boundary¶
Supported after successful execution and analysis:
- calibrated relationship between explicit noise scale and parity survival;
- comparison between standard extrapolation and symmetry-decay guided extrapolation for the selected observable;
- statement that the witness is or is not stable enough for the tested backend/circuit family.
Blocked even after a positive result:
- broad quantum advantage;
- universal error mitigation;
- backend-general GUESS performance;
- DLA-parity-only protection;
- transfer of fitted coefficients to a different backend, layout, or calibration window without new validation.
Circuit Matrix¶
Default scope:
| Field | Value |
|---|---|
n |
4 |
| States | 0011 even reference, 0001 odd reference |
| Depths | 6, 8, 10, 14 |
| Noise scales | 1, 3, 5 via explicit unitary folding or approved equivalent |
| Repetitions | 8 per state/depth/noise-scale |
| Main shots | 4096 per circuit |
| Readout states | 0011, 0001, 0000, 1111 |
| Readout shots | 8192 per circuit |
Circuit count:
- main circuits:
2 states x 4 depths x 3 noise scales x 8 reps = 192; - readout circuits:
4; - total circuits:
196.
Optional extension after the default block:
- add popcount-control states
0101,0010, and0111at depths8and14only if the default block passes the live depth and cost gates.
QPU-Time Estimate¶
Expected IBM-reported QPU time for the default block: 5-12 minutes,
depending on folded circuit depths.
Budget ceiling for default block: 15 IBM-reported QPU minutes.
Optional extension ceiling: additional 15 IBM-reported QPU minutes.
Abort before submission if folded circuits exceed the ceiling or if the remaining allocation cannot cover the block plus a 25 % safety margin.
Live Readiness Gates¶
Before submission:
- confirm selected backend is Heron-class, account-visible, and operational;
- generate unfolded and folded circuits from committed code only;
- live-transpile every folded circuit;
- reject if max depth at noise scale
5exceeds the unscaled depth by more than the preregistered folding expectation plus 25 % overhead; - reject if live two-qubit gate count makes the QPU-time estimate exceed the budget ceiling;
- record backend name, calibration timestamp, folding rule, depth summary, two-qubit gate summary, circuit count, shot count, and estimated QPU minutes;
- get explicit approval immediately before submission.
Analysis Plan¶
Primary witness:
- parity survival and parity leakage versus explicit noise scale.
Primary extrapolation:
- fit log survival or leakage trend versus noise scale for each state/depth;
- compare ordinary extrapolation against symmetry-decay guided extrapolation;
- report calibration coefficients only for the sampled backend/layout/window.
Promoted summaries:
- per-depth witness stability;
- fit quality by state and sector;
- extrapolated observable shift with confidence interval;
- failure modes where folding destroys signal or fit quality.
Readout handling:
- apply exact-state parity readout correction where the four readout states support it;
- do not claim full confusion-matrix mitigation from four-state readout calibration.
Falsification Rules¶
The GUESS witness claim is rejected or downgraded if:
- parity survival is not monotone or fit quality is poor across noise scales;
- folded circuits introduce layout/transpilation artefacts larger than the witness trend;
- even and odd sectors require incompatible witness models;
- readout correction changes the sign or removes the extrapolation signal;
- the fitted coefficients are unstable across depths.
If the result is negative, report it as a boundary on parity leakage as a GUESS witness for this circuit/backend family.
Output Artefacts¶
Expected paths after approved execution:
data/phase3_guess_dla/phase3_guess_<backend>_<timestamp>.json;data/phase3_guess_dla/phase3_guess_summary_<date>.json;data/phase3_guess_dla/phase3_guess_fit_rows_<date>.csv;data/phase3_guess_dla/phase3_guess_extrapolation_rows_<date>.csv;docs/campaigns/phase3_guess_dla_manifest_<date>.md.
Each artefact must include backend, folding rule, circuit metadata, raw counts, SHA256 hashes, depth/gate summaries, fit diagnostics, and reproduction commands.
Submission Boundary¶
This preregistration is complete. QPU execution remains blocked until backend selection, folded-circuit readiness artefacts, budget confirmation, and explicit approval are completed in a separate task.