SZLHOLDINGS/oac-clinical-transport-observability-synthetic

OAC Clinical Transport Observability — Synthetic

This dataset contains 1,200 fixed-seed, entirely synthetic operational transport-health examples for the companion OAC System Health v1 model. It contains no records collected from a patient, laboratory, analyzer, instrument, LIS, EHR, network, or health-care site.

Companion model: OAC System Health v1. Canonical source: szl-forge clinical gateway.

Data boundary

The closed schema contains only eight bounded operational counters/flags and a synthetic operator_attention_required label. It contains no PHI, personal identifiers, patient/order/specimen fields, assay data, observations, diagnostic content, result values, raw HL7, or FHIR resources.

This dataset is unsuitable for medicine, diagnosis, prognosis, treatment, triage, result interpretation, autoverification, result release, device control, or claims about real-world performance. It must not be joined with patient or clinical data.

Splits

Each JSONL row has this shape:

{
  "features": {
    "configuration_valid": 1,
    "consecutive_failures": 0,
    "ledger_integrity_ok": 1,
    "listener_running": 1,
    "peer_allowlist_configured": 1,
    "queue_utilization": 0.12,
    "seconds_since_last_success": 14.0,
    "tls_enabled": 1
  },
  "label": {"operator_attention_required": false},
  "sample_id": "train-000000",
  "schema": "szl-oac/transport-health-observation/v1",
  "synthetic": true
}

schema.json is a closed JSON Schema. dataset_receipt.json records the fixed seed, row counts, generator hash, schema hash, and SHA-256 for every split.

Reproduction

training_source_snapshot.py is the exact generator/trainer source snapshot used for the published artifacts. It is included for inspection and hashing; its canonical source-tree layout also requires src/oac_operational_health.py. Clone the source repository and run:

git clone https://github.com/szl-holdings/szl-forge.git
cd szl-forge/clinical-gateway
python -I -B tools/train_operational_health_model.py --verify

The generator uses Python's standard library only. Synthetic labels are sampled from a documented operational risk function under a fixed pseudorandom seed; they are generated targets, not human annotations or ground truth about a real system.

License

Apache-2.0. See LICENSE.