We keep the 215 K mW liquid frames in the Hugging Face dataset d-SEAMS/mw-liquid-215K.
DVC is the other home for those frames, once a remote is configured.
Zenodo by itself is not that home.

input/mw215/fetch.py downloads the frames from that dataset.
The download uses huggingface_hub and sends no token.
A failed download falls back to input/mw215/regenerate.py.
The velocity seed in the deck is 215.
One OpenMP thread keeps that seed fixed.
The dumps land under data/mw215.

The deck melts an mW lattice at 500 K for 30 ps.
The same run then stays at 215 K for 50 ps.
The number density is 0.033 molecules per cubic angstrom.
One edge of eight cells gives 512 molecules.
An edge of sixteen cells gives 4096 molecules.
An edge of thirty-two cells gives 32768 molecules.

We leave the compressed dumps on the dataset.
Git keeps the deck and the scripts.
Git leaves the dumps themselves out of the history.
