Synthetic data

The included example contains 1,074 measurements for 12 synthetic people. The R simulator creates every value without using patient records, then introduces missing values, implausible measurements, and BMI mismatches so we have concrete problems to work through.

Provenance

The simulator comes from Part 1, revision 34830a3d1323d9f81a395f5d4fda48b5ee80566b (September 5, 2026), with its script path updated for Part 2. Both guides use the same synthetic schemas and install dependencies as a separate setup step. The upstream source and local file hashes let you identify the version used here.

We generated these files with R 4.6.1, optparse 1.8.2, seed 123, and 12 people. The simulator requires R 4.3 or later; use the recorded versions when you want to reproduce the files exactly. You only need R to regenerate data—the included files are ready to use with the Python pipeline.

The demographic age is measured as of December 31, 2019. A person’s age at an earlier measurement can therefore be different, and some measurements concern minors. The data contract explains how this teaching pipeline treats those records.

Regenerate in a separate directory

Install optparse explicitly if your R environment lacks it:

Rscript -e 'install.packages("optparse", repos = "https://cloud.r-project.org")'

For an R project with an existing lockfile, restore that project’s environment instead. The simulator checks dependencies and does not install them during an analysis run.

Rscript scripts/r/simulate_ehr_data.R \
  --output_ehr tmp/regenerated/ehr_bmi_simulated_data.tsv \
  --output_ehr_dict tmp/regenerated/data_dictionary.txt \
  --output_demo tmp/regenerated/demographics_simulated_data.tsv \
  --output_demo_dict tmp/regenerated/demographics_data_dictionary.txt \
  --seed 123 --n_individuals 12
python scripts/py/check_example_data.py --directory tmp/regenerated

The checker compares the four files with the hashes recorded in data/example/provenance.json. If they differ, first compare your R and package versions and the simulator revision with those above. If you intend to update the included example, review the regenerated data and counts with a collaborator, then update the provenance record and this page together.

Expected baseline

Run the quickstart command with the default mismatch threshold of 2.0.

Metric	Expected value
Input measurements	1,074
Demographics people	12
Selected people	12
Excluded measurement rows	218
Retained measurements before selection	856
People with no selected measurement	0
Per-person IQR outliers	11
Rows carrying a BMI mismatch flag	51

A row can have both a height or weight issue and a BMI mismatch, so reason counts can overlap. These totals provide a quick check that the full pipeline still behaves as expected. Smaller test datasets with answers worked out independently check individual calculations and selection rules.

For a larger practice dataset, use the same command with output paths under data/raw/ and a larger --n_individuals. Its counts will differ from this baseline. You can inspect the included files, dictionaries, and provenance in the repository.