This repository is currently a research codebase under active development. The main goal of these guidelines is to preserve reproducibility and avoid mixing source code with generated artifacts.
- keep source code in
code/ - do not commit large generated outputs from
results/orfigures/ - do not commit local data files from
data/ - do not vendor external repositories directly into this repository unless explicitly decided
- run the relevant R parse checks for edited scripts
- confirm that paths are repository-root relative when possible
- keep benchmark logic and manuscript-facing scripts clearly named
- prefer text documentation over binary notes or ad hoc scratch files
- model definition scripts:
ModelName.R - benchmark runners:
run_*.R - one-off benchmark summaries should live under
results/, not in the repository root
When you add a new model family or benchmark runner, document:
- what question it answers
- what benchmark or validation protocol it uses
- what outputs it writes and where
- whether it is exploratory, confirmatory, or manuscript-facing
As stable components emerge, they should be migrated from code/ into package-style structure. The roadmap is described in docs/package-roadmap.md.