Domain Application Honesty Kits

Domain-facing examples are useful only when their evidence boundary survives copying into notebooks, proposals, and downstream integrations. BL-63 makes that boundary executable. Each honesty kit states what the software route can demonstrate, what data provenance it admits, and which claims remain forbidden.

The kits are policy-bearing software records, not certificates of scientific or operational validity. They do not promote a synthetic fixture, curated matrix, structural proxy, or passing unit test into domain evidence.

Quick start

from scpn_quantum_control.applications import (
    build_application_honesty_audit_report,
    get_domain_application_honesty_kit,
)

kit = get_domain_application_honesty_kit("eeg_like_synthetic")
assert kit.synthetic_only
assert not kit.publication_safe
assert "clinical classification" in kit.claims_forbidden

report = build_application_honesty_audit_report()
assert report.passed
print(report.content_digest())

This code reads only versioned packaged application artifacts. It performs no network access, private-data discovery, provider submission, or hardware work.

Policy schema

DomainApplicationHonestyKit is immutable. Its fields have distinct jobs:

Field	Meaning
kit_id	Stable machine identifier used for lookup and evidence.
domain_tag	Non-promotional label for reports and interfaces.
support_status	bounded_research or simulation_only.
data_origin	synthetic or curated_public.
synthetic_only	Exact shorthand that must agree with data_origin.
dataset_ids	Packaged public artifacts governed by the kit; forbidden for synthetic-only kits.
source_modules	Implementing import paths, not a claim of domain fidelity.
allowed_uses	Narrow software demonstrations admitted by the kit.
caveats	Scientific and operational limitations that must travel with the result.
claims_forbidden	Explicit claims the kit never authorises.
forecasting_tags	BL-37 generator tags; always simulation-only.
publication_safe	Always False: the kit itself is not domain-publication evidence.

Construction fails closed on blank or duplicate policy text, enum bypasses, inconsistent synthetic flags, packaged datasets on synthetic-only routes, and duplicate forecasting tags.

Built-in kits

Public power-grid benchmark

power_grid_public_benchmark governs the packaged IEEE 5-bus constants. It supports artifact-custody checks, Kuramoto compilation, and structural-proxy software tests. It does not support live-grid control, transient-stability prediction, utility deployment, or quantum-advantage claims.

The related BL-37 tag is grid_like_sim. That tag still denotes a generated forecasting configuration, even though this separate kit also governs one curated public benchmark.

Illustrative Josephson simulation

josephson_illustrative_simulation admits generated coupling topologies and explicitly labelled nominal literature parameters. It is simulation-only. An XY structural analogy is not measured-device validation, hardware self-simulation, calibration guidance, or fabrication evidence.

Synthetic EEG-like software route

eeg_like_synthetic admits generated PLV-shaped inputs and the BL-37 eeg_like_sim tag. It deliberately names no packaged dataset. EEG-like means only that a generator has a documented matrix shape and feature convention; it does not mean clinical classification, neural-dynamics reproduction, diagnosis, therapy, or human-subject generalisation.

The separately packaged eeg_alpha_plv_8ch artifact remains a small curated public-literature benchmark matrix. It is audited by the dataset privacy API, but it is not silently relabelled as synthetic or promoted by this kit.

ITER-inspired disruption simulation

iter_disruption_inspired_simulation governs opt-in generated disruption features and local advisory-contract exercises. Its related BL-37 tag is plasma_like_sim. It does not validate ITER prediction, facility operation, real-time controller admission, or plasma-operation guidance.

Measured fusion archives remain outside this kit and require their own custody, licence, validation, and downstream control-admission gates.

Dataset privacy audit

audit_application_benchmark_privacy() validates every artifact in data/public_application_benchmarks/ against its descriptor:

• stable source identity and domain;
• exact curated source mode;
• an exact privacy/licence boundary embedded in artifact metadata;
• contains_personal_data=False for the packaged bytes;
• publication-safe artifact custody and array hashes.

The audit covers EEG, plasma, power-grid, and FEP packaged rows. It does not claim that arbitrary third-party plugin inputs are public or safe. Plugin authors remain responsible for their own raw-data governance.

from scpn_quantum_control.applications import audit_application_benchmark_privacy

for row in audit_application_benchmark_privacy():
    print(row.dataset_id, row.privacy_classification, row.passed)

A mismatch raises immediately. The API does not return an ambiguous "mostly passed" catalogue.

Deterministic evidence

Regenerate the committed report with:

PYTHONPATH=src:oscillatools/src .venv-linux/bin/python scripts/run_application_honesty_audit.py

Verify byte-for-byte freshness without writing:

PYTHONPATH=src:oscillatools/src .venv-linux/bin/python scripts/run_application_honesty_audit.py --check

The JSON payload includes the full kits, privacy rows, artifact hashes, global claim boundary, and a SHA-256 content digest. The Markdown file is a compact human-readable view of the same report. Neither file is measured-domain, clinical, facility, hardware, or advantage evidence.

Extending the registry

A new kit should be added only with all of the following:

1. one stable, non-promotional identifier;
2. an exact data-origin decision;
3. source modules and any governed packaged dataset identifiers;
4. positive allowed uses plus caveats and forbidden claims;
5. privacy metadata for every new packaged artifact;
6. focused tests for failure paths, JSON stability, and evidence freshness;
7. documentation and generated capability inventory updates.

Do not reuse a simulation tag for measured data. Do not mark curated data as synthetic. Do not use publication_safe=False as a substitute for concrete forbidden claims. Provider, QPU, clinical, facility, operational, and advantage promotion remain separate evidence programmes.

Complete API reference

The Application Honesty API documents every public enum, record, lookup, audit, digest, and renderer directly from source docstrings.