Readout Full-Basis Manifest¶

Date: 2026-05-06

This manifest records the full computational-basis readout calibration for the Phase 3 ibm_marrakesh DLA replication dataset.

Scope¶

Target dataset:

data/phase3_multidevice_dla/phase3_multidevice_ibm_marrakesh_2026-05-06T171231Z.json

Execution script:

scripts/readout_full_basis_calibration_ibm.py

Backend and layout:

Field	Value
Backend	`ibm_marrakesh`
Physical qubits	`[5, 6, 7, 8]`
Logical qubits	`4`
Basis states	`16`
Shots per state	`8192`
Job ID	`ibm-run-ddd29a2fbcaeed61`
Estimated QPU minutes	`0.14666666666666667`
Preregistered ceiling	`5` minutes

Live readiness:

depth min/mean/max: 1 / 1.9375 / 2
gate min/mean/max: 4 / 6.0 / 8

Artefacts¶

Artefact	SHA256
`data/readout_full_basis/readout_full_basis_ibm_marrakesh_4q_2026-05-06T173054Z.json`	`ad49289a204c2a54c31ace88ac903bbf6fa8a1afe869b54e691ee1808c4f92b1`
`data/readout_full_basis/readout_full_basis_summary_2026-05-06.json`	`315aefef880a8c4bf328a20e7ec99eacdfe68db81b9dff9db8aec94c06fac8ac`
`data/readout_full_basis/readout_full_basis_matrix_2026-05-06.csv`	`897748bedea05b3ed2aef71342d01d8e5f6823d902a3a23e6e0a279715a108bf`
`data/readout_full_basis/readout_full_basis_rows_2026-05-06.csv`	`0ae1b26c6db1d0e0a5303ae8589a1c54f959be8fbd82c3436a03e4f3a16fd4a7`

Calibration Quality¶

Metric	Value
Mean retention	`0.9699935913085938`
Minimum retention	`0.952392578125`
Maximum retention	`0.98193359375`
Mean parity-flip rate	`0.0296478271484375`
Maximum parity-flip rate	`0.0472412109375`
Assignment-matrix condition number	`1.0756988057943142`

The condition number is low, so the assignment matrix is numerically stable for readout-only correction of the matching ibm_marrakesh four-qubit layout.

Claim Boundary¶

Supported:

full-basis readout assignment calibration for ibm_marrakesh physical qubits [5,6,7,8];
readout-only correction eligibility for the matching Phase 3 DLA raw-count dataset;
row retention, parity-flip, and condition-number diagnostics.

Blocked:

mitigation of gate errors, Trotter error, decoherence, or crosstalk;
reuse on another backend, another physical layout, or a distant calibration window without drift justification;
any claim that readout calibration alone explains the Phase 3 sign pattern until the DLA rows are explicitly reanalysed with this matrix.

Reproduction¶

The submitted command was:

./.venv-linux/bin/python scripts/readout_full_basis_calibration_ibm.py \
  --backend ibm_marrakesh \
  --physical-qubits 5,6,7,8 \
  --submit \
  --confirm-budget

No further QPU work is required to rebuild the matrix and summary artefacts from the raw-count JSON.