Analysis code and data for:
What’s here
analysis/ numbered pipeline scripts (01_clean.py ... 04_check.py)
data/ raw data (or fetch instructions; see below)
results/ regenerated by the pipeline; not committed
figures/ regenerated by the pipeline; not committed
pyproject.toml + uv.lock exact software environment
Makefile one-command runner (`make all`)
Reproduce the results
Requires: uv >= 0.7, GNU make. Tested on: macOS 15.5 and Ubuntu 24.04, ~4 minutes, < 2 GB RAM.
git clone <repo-url> && cd <repo>
make all # regenerates results/ and figures/ from raw data
make check # asserts every number claimed in the paper
Expected final line: all published values reproduced
Data
data/raw.csv:. SHA-256: ` `. - <If restricted: exactly who can access it, how to request it, and what synthetic/subset stand-in is included so the pipeline still runs.>
Randomness and tolerances
All analyses seed their generators (seeds recorded in the scripts).
Stochastic results (bootstrap CIs) are bit-exact given the seed and stable
to +/-0.05 across seeds; make check enforces the reported tolerances.
Environment notes
Python 3.12.3, dependencies locked in uv.lock. A container image is
archived with the release for long-term execution:
docker load < rt-study-v1-image.tar.gz.
License and citation
Code: MIT. Data: