Reproduce the reference and independent studies

The reproduction/ folder in the GitHub repository contains software and run definitions. It has no saved performance reports. Download the repository and run commands from its reproduction directory:

cd reproduction
python3 verify_sources.py
Rscript --vanilla RUNME.R --help
Rscript --vanilla RUNME.R install --study reference22
Rscript --vanilla RUNME.R install --study independent192
Rscript --vanilla RUNME.R check --study reference22
Rscript --vanilla RUNME.R check --study independent192

The runners install archived research packages into separate project libraries. They do not download third-party dependencies. A local source fingerprint records which engine is used. RUN_GUIDE.txt lists the declared dependencies, recorded versions and explicit execution/resume/verification commands.

Study	Scenarios	Replications per scenario	Engine
Reference	22	100	Original R softridge 0.4.0
Independent	192	500	RcppArmadillo softridge 0.3.0.1 + scaleCorrection 0.1.0

The studies retain their own seeds, designs and tuning protocols. The 12-by-40 reference grid differs from the independent 20-by-41 fold-local grid. The different version numbers identify historical implementations; they do not imply that the pure-R reference supersedes the compiled research engine.

Only explicit --all --confirm-full commands start complete studies. The independent runner has ten validated batch checkpoints, a single writer, signature-checked resume and checksum verification. Existing outputs are protected. End-to-end portable full-run cost has not been measured; inspect a bounded example before planning an independent full reproduction.

Data dictionary

Input	Meaning
config.rds	Frozen numerical study definition and seed list
scenarios.csv	Sample size, dimension, sparsity, AR correlation and SNR
tasks.csv	Scenario-replication identities and recorded order
signals.csv	Active-variable locations and coefficients
protocol_corrected.json	Corrected-scale protocol and metric definitions

The optional riboflavin example uses public hdi data. It is downloaded only through the explicit fetch_riboflavin.R command and checked against a pinned SHA-256 digest. Original data are retained without modification.