Skip to content

SPDX-License-Identifier: AGPL-3.0-or-later

Commercial license available

© Concepts 1996–2026 Miroslav Šotek. All rights reserved.

© Code 2020–2026 Miroslav Šotek. All rights reserved.

ORCID: 0009-0009-3560-0851

Contact: www.anulum.li | protoscience@anulum.li

SCPN Quantum Control — SCPN/FIM Validation Protocol

SCPN/FIM Validation Protocol

Date: 2026-05-05

Principle

No SCPN/FIM Hamiltonian paper claim is accepted from memory, previous applications, or informal notes. A claim becomes usable only when it is backed by a committed script, a committed JSON/CSV artefact, and a documented claim boundary.

Phase 0 — Claim inventory

Goal: separate regenerated claims from historical or speculative claims.

Deliverables:

  • Table of candidate claims.
  • Status for each claim: validated, prediction, historical-unverified, or rejected.
  • Source artefact path for every validated or predicted claim.

Hard rule:

  • Historical IBM-credit-application numbers are not manuscript evidence until regenerated.

Phase 1 — Offline Hamiltonian validation

Goal: establish what the FIM term mathematically does before noise, circuits, or hardware.

Initial implementation status:

  • Exact spectrum and magnetisation-sector summaries are implemented in scripts/analyse_fim_spectrum.py.
  • Adjacent-gap-ratio summaries are implemented in scripts/analyse_fim_level_spacing.py.
  • Bipartition-entropy summaries are implemented in scripts/analyse_fim_entanglement.py.
  • Ideal sector-conservation and energy-barrier summaries are implemented in scripts/analyse_fim_sector_survival.py.
  • Small-n VQE ground-state summaries are implemented in scripts/benchmark_fim_vqe_ground_state.py.
  • A non-submitting IBM pilot candidate protocol is implemented in scripts/prepare_fim_ibm_pilot.py.
  • Initial n=4,6,8 artefacts were generated on 2026-05-05 under data/scpn_fim_hamiltonian/.

Required outputs:

  • Spectrum versus lambda.
  • Sector energies versus magnetisation.
  • Gap statistics.
  • Level-spacing ratio.
  • Eigenstate magnetisation distribution.
  • Entanglement entropy where feasible.

Candidate parameter grid:

  • n in {4, 6, 8, 10, 12} where feasible.
  • lambda in {0, 0.25, 0.5, 1, 2, 4, 8} unless offline scans suggest a better scale.
  • Fixed heterogeneous Kuramoto coupling matrix with recorded K_ij parameters.

Exit criterion:

  • At least one effect is large, reproducible, and interpretable enough to design a falsification experiment.

Phase 2 — Noise and circuit feasibility

Goal: determine whether any Phase 1 effect is likely to survive NISQ execution.

Required checks:

  • Equal-depth state preparation.
  • Transpiled two-qubit gate counts.
  • Layout sensitivity.
  • Popcount and magnetisation confound separation.
  • Readout-only baseline requirement.
  • Classical noise-model survival prediction.

Scientific boundary:

  • The ideal Hamiltonian conserves total magnetisation, so the offline sector-survival artefact has zero ideal unitary sector leakage by construction.
  • IBM leakage or survival asymmetry is therefore not a direct ideal-Hamiltonian effect. It must be modelled and tested as open-system or circuit-execution behaviour.

Exit criterion:

  • A minimal n=4 IBM pilot protocol exists with predicted effect size, shot budget, circuit count, backend assumptions, and stop rule.

Phase 3 — IBM pilot gate

Goal: spend only the QPU time needed to falsify the strongest offline prediction.

Pilot constraints:

  • n=4 first.
  • Compare lambda = 0 against selected non-zero lambda values.
  • Use equal-depth circuits.
  • Include magnetisation/popcount controls.
  • Use same-day calibration metadata.
  • Record raw counts, backend, layout, transpilation seed, circuit depth, shot count, and job IDs.

Current candidate protocol:

  • lambda in {0, 1, 4}.
  • depths {2, 4, 6}.
  • five representative magnetisation sectors with explicit popcount metadata.
  • full 16-basis-state readout baseline.
  • 4096 shots per candidate circuit.
  • 61 total candidate circuits and 249856 total candidate shots.
  • submission status: not_submitted.

Local circuit-preparation status:

  • Generated by scripts/prepare_fim_ibm_circuits.py.
  • Artefacts: fim_ibm_circuit_preparation_2026-05-05.json and .csv.
  • Local basis-gate transpilation only; not backend-calibrated Heron ISA.
  • 61 circuits and 249856 total shots.
  • Maximum local transpiled depth: 262.
  • Maximum local transpiled two-qubit gate count: 144.
  • Submission status: not_submitted.

Live IBM readiness status:

  • Script added: scripts/prepare_fim_ibm_live_readiness.py.
  • Scoped lint passed.
  • Live backend authentication and backend selection succeeded after matching the vault parser to the legacy IBM script format.
  • Backend: ibm_kingston.
  • Artefacts: fim_ibm_live_readiness_2026-05-05.json and .csv.
  • Total circuits: 61.
  • Total shots: 249856.
  • Maximum live transpiled depth: 540.
  • Maximum live transpiled two-qubit gate count: 157.
  • No Sampler, Estimator, Runtime session, or QPU job was submitted.
  • Submission status: not_submitted.

Pilot outcomes:

  • confirmed_direction: observed sign agrees with offline prediction within the pre-registered analysis boundary.
  • falsified_direction: observed sign contradicts offline prediction.
  • hardware_inconclusive: uncertainty, drift, readout, or layout sensitivity is too large for a claim.

Submission status:

  • Submitted after live readiness approval on 2026-05-05.
  • Backend: ibm_kingston.
  • Public run label: ibm-run-4c0bd60c3fc2c532.
  • Circuits: 61.
  • Shots: 249856 total.
  • Raw-count artefact: data/scpn_fim_hamiltonian/fim_ibm_pilot_raw_counts_2026-05-05_ibm-run-4c0bd60c3fc2c532.json.
  • Raw-count SHA256: be284b9b2f71dfecd978703d979a8893e79b35dcc4537d7a372b83ba48305790.
  • Analysis status: pending. No hardware claim is promoted until analysis and readout controls are computed.

Phase 4 — Paper integration

Allowed paper language depends on phase:

  • Phase 1 only: theoretical/computational paper with hardware-facing predictions.
  • Phase 2 complete: computational paper with feasible, pre-registered IBM protocol.
  • Phase 3 complete: hardware-supported paper if the pilot is conclusive.

Required claim table:

The current claim table is maintained in docs/scpn_fim_claim_boundary_2026-05-05.md.

Phase 5 — Adaptive lambda feedback loop

Goal: turn the fixed H_FIM(lambda) study into a falsifiable closed-loop protocol without claiming hardware validation before a new run exists.

Implemented scaffold:

  • Module: src/scpn_quantum_control/analysis/adaptive_fim_feedback.py.
  • Public API: AdaptiveFIMConfig, FIMWitness, propose_next_lambda, and adaptive_lambda_schedule.
  • Tests: tests/test_adaptive_fim_feedback.py.

Initial control rule:

  • leakage_suppression: reduce lambda when measured leakage exceeds the target witness.
  • retention_recovery: reduce lambda when exact-state retention falls below the target witness.
  • Updates are clipped to a configured [lambda_min, lambda_max] interval and support a deadband so small shot-noise fluctuations do not change the next batch.

Scientific boundary:

  • This is a deterministic classical controller over simulator or already measured witness values.
  • It does not submit QPU jobs.
  • It does not reverse the negative fixed-lambda IBM result.
  • A hardware claim would require a separately approved adaptive or batched follow-up experiment with fresh raw counts, live transpilation metadata, readout controls, and a documented stop rule.

Phase 6 — Full-basis readout-matrix mitigation cross-check

Goal: close the scalable readout-mitigation review point without spending new QPU time.

Implemented path:

  • Module: src/scpn_quantum_control/mitigation/readout_matrix.py.
  • Script: scripts/analyse_fim_readout_matrix_mitigation.py.
  • CLI wiring: scpn-bench fim-all --include-readout.

Scientific boundary:

  • The repeated SCPN/FIM follow-up includes all 16 computational-basis readout calibration circuits for n=4, so a literal 2^n x 2^n confusion-matrix inversion is valid for that dataset.
  • The mitigation operates only on measurement confusion in the committed raw counts. It does not correct coherent gate errors, Trotter overhead, layout routing noise, or backend drift.
  • Datasets without a complete basis calibration remain limited to exact-state or parity-readout corrections unless new calibration circuits are collected.

QPU budget guard

No IBM job should be submitted until:

  • Offline artefacts exist.
  • Circuits are generated and counted.
  • The expected QPU time is estimated.
  • The falsification criterion is written down.
  • The user explicitly approves the run.

2026-05-05 IBM pilot analysis result

The n=4 SCPN/FIM IBM pilot was submitted to ibm_kingston as job ibm-run-4c0bd60c3fc2c532 and returned complete raw counts for 61 circuits and 249,856 shots.

Analysis artefacts:

  • scripts/analyse_fim_ibm_pilot.py
  • data/scpn_fim_hamiltonian/fim_ibm_pilot_analysis_2026-05-05_ibm-run-4c0bd60c3fc2c532.json
  • data/scpn_fim_hamiltonian/fim_ibm_pilot_row_metrics_2026-05-05_ibm-run-4c0bd60c3fc2c532.csv
  • data/scpn_fim_hamiltonian/fim_ibm_pilot_lambda_trends_2026-05-05_ibm-run-4c0bd60c3fc2c532.csv
  • docs/scpn_fim_ibm_pilot_analysis_2026-05-05.md

Outcome: descriptive pilot only. The run does not support a hardware claim that the FIM term improves coherence. For lambda = 4 relative to lambda = 0, the mean magnetisation-leakage delta was +0.11559244791666666 across 15 matched state/depth comparisons, with higher magnetisation leakage in 15 / 15 comparisons. Because the pilot has one sample per lambda/depth/state condition, no formal p-values are reported.