1,000,000 synthetic NHS-style patient records across three tiers of richness. Built for search benchmarking, DMM ingest testing, and developer validation of health-data pipelines. No real patient data. No real identifiers.

Dataset Summary

Lindisfarne M1 is a fully synthetic corpus of one million NHS-style health records generated by the Toridion TQNN DMM synthetic data pipeline. Records are modelled on real NHS clinical data structures — SNOMED CT, HL7 v2.5, FHIR R4, and a GHX variant — but contain no real patient information of any kind. Every name, NHS number, NI number, address, postcode, diagnosis, medication, and clinician is procedurally generated.

The dataset is published in three tiers to serve different use cases:

Tier	Files	Fields per record	Description
Full	`full/`	20–35	Rich nested records, all formats, labs, allergies, alerts, procedures
Slim	`slim/`	10–15	Reduced nesting, max 2 narrative strings, faster ingest
Superslim	`superslim/`	10 (flat)	Flat JSON, no arrays, maximum ingest speed

All three tiers contain exactly 1,000,000 records in JSONL format, split into chunks of 10,000 records per file (100 files per tier).

Intended Use

Search benchmarking — test retrieval systems against a realistic, large-scale health record corpus at known scale (1M records, three richness levels)
DMM ingest testing — validate document memory and storage pipelines including the Toridion TQNN DMM appliance
Developer validation — build and test health-data ETL pipelines, parsers, indexers, and explorers against plausible but entirely safe synthetic data
ML/AI research — pre-training, fine-tuning, or evaluation on structured clinical text without any patient privacy risk

Out of Scope

This dataset is not suitable for clinical decision support, medical research, or any application that requires statistically representative real-world health data. The distributions (diagnosis prevalence, medication frequency, demographic mix) are plausible but not calibrated to real NHS statistics.

Record Formats

Records are distributed across four clinical message formats in approximately the following proportions (consistent across all three tiers):

Format	Approx. Share	Description
HL7 v2.5	34%	ADT^A01 admission messages with PID, PV1, DG1, RX1, OBX, AL1 segments
FHIR R4	34%	Patient / Encounter / Condition / MedicationRequest / Observation bundles
SNOMED CT	28%	Coded records with SNOMED diagnosis, medication, lab, and procedure codes
GHX	4%	Generic Health Exchange variant

Field Schemas

Superslim (10 flat fields)

{
  "id":    "LM1-000001",
  "name":  "Oliver Smith",
  "nhs":   "485 621 3947",
  "email": "o.smith@nhs-patient.example",
  "dob":   "1978-03-14",
  "visit": "2024-07-22",
  "diag":  "44054006",
  "dlbl":  "Diabetes mellitus type 2",
  "med":   "372756006",
  "fmt":   "SNOMED"
}

Slim — SNOMED example

{
  "record_id": "LM1-000001",
  "patient_info": {
    "full_name":  "Oliver Smith",
    "dob":        "1978-03-14",
    "nhs_number": "485 621 3947",
    "phone":      "07700 900142",
    "email":      "o.smith@nhs-patient.example",
    "gender":     "Male",
    "gp_surgery": "Parkway Medical Centre"
  },
  "clinical_encounter": {
    "date":      "2024-07-22",
    "type":      "Outpatient",
    "clinician": "Dr Sarah Patel"
  },
  "SNOMED_codes": {
    "diagnosis":  "44054006",
    "diag_label": "Diabetes mellitus type 2",
    "medication": "372756006",
    "lab":        "Fasting glucose",
    "procedure":  "73761001"
  },
  "format": "SNOMED"
}

Full — HL7 example (abbreviated)

{
  "HL7MessageID": "LM1-000002",
  "MSH": {
    "SendingFacility":   "Newcastle upon Tyne Hospitals NHS Foundation Trust",
    "ReceivingFacility": "NHS Spine",
    "MessageDateTime":   "2024-02-15T09:32:11",
    "MessageType":       "ADT^A01",
    "Version":           "2.5"
  },
  "PID": {
    "PatientName": "Amelia Khan",
    "DateOfBirth": "1991-11-02",
    "NHSNumber":   "712 334 8820",
    "NI_Number":   "JT 84 21 97 C",
    "Gender":      "Female",
    "Address": {
      "line1":    "47 Tyne View",
      "town":     "Gateshead",
      "postcode": "NE8 3AB",
      "country":  "England"
    },
    "Phone": "07700 900381",
    "Email": "a.khan@nhs-patient.example"
  },
  "PV1": {
    "VisitType":       "Inpatient",
    "VisitDate":       "2024-02-15",
    "Ward":            "Ward C4",
    "Consultant":      "Prof Mohammed Williams",
    "ConsultantEmail": "m.williams@newcastle-hospitals.nhs.uk",
    "AdmitSource":     "GP Referral"
  },
  "DG1": {
    "DiagnosisCode": "I10",
    "DiagnosisDesc": "Essential hypertension",
    "DiagnosisType": "Primary"
  },
  "RX1": [{
    "MedicationCode": "C03CA01",
    "MedicationName": "Furosemide",
    "Dose":           "40mg",
    "Frequency":      "Once daily",
    "Route":          "Oral",
    "PrescribedBy":   "Dr Emily Brown"
  }],
  "OBX": [{
    "test":  "Blood pressure",
    "value": "142/91 mmHg",
    "date":  "2024-02-15",
    "flag":  "H"
  }],
  "AL1": [{
    "allergen":  "Penicillin",
    "reaction":  "Rash",
    "severity":  "Moderate",
    "confirmed": "Yes"
  }],
  "format": "HL7"
}

Data Generation

Records were generated using the Toridion TQNN DMM Synthetic Health Record Generator.

Key generation properties:

Names — drawn from a diverse forename/surname pool covering English, South Asian, East Asian, Middle Eastern, Irish, Scottish, and West African naming conventions, reflecting the demographic diversity of NHS patient populations
NHS numbers — randomly formatted 10-digit strings conforming to the NHS number display format (NNN NNN NNNN); not valid NHS numbers
NI numbers — randomly formatted strings conforming to UK NI display conventions; not real NI numbers
Addresses — street names, towns, and postcode prefixes drawn from North East England (NE, SR, DH, TS, DL prefix areas); postcodes are randomly generated and do not correspond to real addresses
NHS Trusts — real trust names used as plausible sending facilities in HL7/FHIR records; no real patient data is associated with these organisations
Diagnoses — SNOMED CT codes and ICD-10 codes drawn from a representative set of common conditions; prevalence is not calibrated to real epidemiological data
Medications — ATC/SNOMED medication codes; dosing is plausible but not clinically validated
Dates — encounter dates distributed randomly across 2020–2026

The generator supports reproducible output via --seed for exact dataset regeneration.

File Structure

Toridion/lindisfarne-m1/
├── README.md
├── full/
│   ├── lindisfarne_full_0001.jsonl      # records 1–10,000
│   ├── lindisfarne_full_0002.jsonl      # records 10,001–20,000
│   ├── ...
│   └── lindisfarne_full_0100.jsonl      # records 990,001–1,000,000
├── slim/
│   ├── lindisfarne_slim_0001.jsonl
│   ├── ...
│   └── lindisfarne_slim_0100.jsonl
└── superslim/
    ├── lindisfarne_superslim_0001.jsonl
    ├── ...
    └── lindisfarne_superslim_0100.jsonl

Each .jsonl file contains 10,000 records, one JSON object per line. 100 files per tier · 3 tiers · 1,000,000 records each.

Loading the Dataset

Python — Hugging Face `datasets` library

from datasets import load_dataset

# Load the superslim tier (fastest, smallest)
ds = load_dataset("Toridion/lindisfarne-m1", data_dir="superslim", split="train")

# Load a single chunk directly
ds = load_dataset(
    "Toridion/lindisfarne-m1",
    data_files="superslim/lindisfarne_superslim_0001.jsonl",
    split="train"
)

# Stream the full tier without downloading everything
ds = load_dataset(
    "Toridion/lindisfarne-m1",
    data_dir="full",
    streaming=True,
    split="train"
)
for record in ds.take(5):
    print(record)

Direct download

# Single chunk
wget https://huggingface.co/datasets/Toridion/lindisfarne-m1/resolve/main/superslim/lindisfarne_superslim_0001.jsonl

# All superslim chunks
huggingface-cli download Toridion/lindisfarne-m1 --repo-type dataset --include "superslim/*"

# Full dataset
huggingface-cli download Toridion/lindisfarne-m1 --repo-type dataset

TQNN DMM ingest

Stream any downloaded JSONL chunk directly into a TQNN DMM appliance endpoint:

# Bash — ingest a chunk line by line
while IFS= read -r line; do
  curl -s -X POST https://your-tqnn-appliance/storeDoc \
    -H "Content-Type: application/json" \
    -d "$line" > /dev/null
done < superslim/lindisfarne_superslim_0001.jsonl

# Python — stream and ingest with progress reporting
import json, requests

endpoint = "https://your-tqnn-appliance/storeDoc"

with open("superslim/lindisfarne_superslim_0001.jsonl") as f:
    for i, line in enumerate(f, 1):
        record = json.loads(line)
        requests.post(endpoint, json=record)
        if i % 1000 == 0:
            print(f"Ingested {i:,} records")

Privacy & Safety

This dataset contains no real patient data. All records are entirely synthetic:

No real NHS numbers, NI numbers, names, addresses, or contact details
No data derived from, seeded by, or statistically calibrated against real patient records
No personally identifiable information of any kind
Safe for unrestricted public distribution, commercial use, and open research

Use of real NHS trust names and clinical code standards (SNOMED CT, HL7, FHIR) is purely for structural realism. Toridion has no affiliation with NHS England or any NHS organisation.

Citation

If you use Lindisfarne M1 in your research or tooling, please cite:

@dataset{toridion_lindisfarne_m1_2026,
  author    = {Toridion},
  title     = {Lindisfarne M1: Synthetic NHS Health Record Corpus},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/Toridion/lindisfarne-m1},
  license   = {CC BY 4.0}
}

About Toridion

Toridion develops the TQNN DMM (Tensor Quantum Neural Network Document Memory Module) — a sovereign, quantum-resistant document memory and search platform designed for deployment at the edge and in secure data centre environments.

Lindisfarne M1 is the primary demo corpus for the TQNN DMM platform, showcasing on-the-fly secure search across large-scale health record corpora without exposing real patient data.

Platform: toridion.com

Licence

Creative Commons Attribution 4.0 International (CC BY 4.0)

You are free to share and adapt this dataset for any purpose, including commercially, provided you give appropriate credit to Toridion.

Downloads last month: 257

Total file size:

1.9 GB

Lindisfarne M1 — Synthetic NHS Health Record Corpus