Data directory notes¶

This page is the English entry point for the repository data directory note.

What the data directory note explains¶

It explains what belongs under data/, why run outputs are grouped by run_id, and why the project avoids scattering long-lived artifacts across unrelated folders.

Why this matters¶

The project is designed for restartable training and later analysis. Both depend on having one predictable place to find run outputs.

Inputs and outputs¶

input: generated checkpoints, game records, metrics, plots, and related run files
output: a consistent data/runs/<run_id>/ structure

How to judge a normal result¶

one run maps to one run directory
top-level data/ does not grow ad hoc subdirectories for each artifact type
tools can locate run outputs without path guessing

Source document¶

The root data directory note remains the authority:

data/README.md