anon-pcqnp-ed26/pcqnp-finite-shot-reliability-artifact

PC-QNP Finite-Shot Reliability Artifact

This anonymous review artifact supports a NeurIPS 2026 Evaluations & Datasets submission on physics-conformal reliability evaluation for finite-shot quantum-process surrogate models. The asset is intended for reviewer inspection and reviewer-level reproduction of the reported aggregate tables, nested residual-repair calculation, and IBM stochastic Pauli-channel protocol validation.

Contents

• —code_snapshot/: cleaned source-code snapshot, selected configurations, and paper-relevant scripts for method inspection and lightweight reruns.
• —data/original_tables/: compact aggregate tables used for the main reliability audit and boundary stress-test claims.
• —data/raw_point_predictions/: raw point-prediction CSVs used to recompute the nested PC-QNP residual-repair result from prediction rows.
• —data/unified_repair/: nested residual-repair summaries, per-bin metrics, fitted gate JSON, and coverage/baseline comparison figures.
• —data/hardware_channel/: compact analysed IBM stochastic Pauli-channel validation summaries.
• —supplement/hardware_channel_validation/: protocol CSVs, raw counts JSON, job/run metadata, and analysed hardware-channel outputs.
• —scripts/recompute_nested_residual_repair.py: standalone script documenting the disjoint gate-training, conformal-calibration, and held-out evaluation split used in the residual-repair evidence.
• —ASSET_CARD_PCQNP.md: human-readable asset card, scope, limitations, and reviewer workflow.
• —REPRODUCIBILITY_COMMANDS.md: command-level instructions for the contained reproduction checks.
• —requirements.txt: lightweight Python requirements for inspecting and recomputing the provided summaries.

Minimal reviewer workflow

1. 1.Read the main submitted paper PDF in OpenReview.
2. 2.Inspect REPRODUCIBILITY_COMMANDS.md for command-level reproduction instructions.
3. 3.Inspect data/unified_repair/unified_residual_repair_summary.csv and data/unified_repair/nested_conformal_baseline_comparison.csv for the nested repair metrics.
4. 4.Recompute the nested PC-QNP residual-repair result from the included raw point-prediction CSVs using scripts/recompute_nested_residual_repair.py.
5. 5.Inspect supplement/hardware_channel_validation/hardware_channel_summary_metrics.csv and supplement/hardware_channel_validation/hardware_channel_metadata.json for the IBM stochastic Pauli-channel validation.
6. 6.Re-run the IBM stochastic Pauli-channel analysis from the included raw counts using code_snapshot/scripts/pcqnp_hardware_channel_validation.py.

Scope and limitations

This artifact supports a reliability/evaluation paper. It does not claim quantum advantage.

The artifact supports direct recomputation of:

1. 1.the nested physics-residual conformal repair from included raw point-prediction CSVs; and
2. 2.the IBM stochastic Pauli-channel validation from included raw count records.

The artifact includes a cleaned source-code snapshot for method inspection and lightweight reruns, but it is not intended to be a bitwise historical reconstruction of every exploratory development run. Full from-scratch retraining of all historical phase-1-to-phase-6 experiments would require the original training workspace, selected checkpoints, exact hardware/GPU details, and a pinned containerized environment.

The IBM hardware evidence validates realised stochastic Pauli-mixture protocol semantics, measurement convention, realised-mixture aggregation, and finite-shot bookkeeping. It does not validate arbitrary noisy-channel surrogate learning or hardware learning of the synthetic amplitude-damping task family.

Responsible AI / data documentation

The data are synthetic simulator outputs, analysis tables, and cloud quantum hardware counts. They do not contain human-subject data, scraped personal data, medical data, demographic attributes, or natural-language user data.

The main risk is scientific overclaiming. The paper and artifact therefore explicitly restrict the supported claims: this is an evaluation-and-repair protocol for finite-shot quantum-process surrogate reliability, not evidence of quantum advantage, scalable quantum superiority, arbitrary-shift conformal validity, or exact conditional coverage.

The file croissant_rai_fields_to_add.json documents Responsible AI metadata fields used for the submission metadata.

License

Data and documentation in this review artifact are released under CC BY 4.0 unless otherwise noted. Reuse should cite the associated paper if accepted.

Reproducibility commands

For exact command-line instructions, including the contained IBM raw-count reanalysis, nested residual-repair recomputation, and paper-relevant phase scripts, see `REPRODUCIBILITY_COMMANDS.md`.

This upload is intentionally a clean reviewer artifact rather than the raw development workspace. It supports reviewer-level verification of the paper's reported hardware-validation and residual-repair results directly from included files.