# Output Artifacts

The standard artifact-writing path produces a small set of files that
capture the model surface, projected compounds, and run metadata.

## CSV files

- `train_coords.csv`
  - embedded training coordinates and labels
- `test_coords.csv`
  - projected test coordinates and scores
- `predictions_test.csv`
  - same tabular test prediction payload used for downstream inspection

## Array artifact

- `surface_grid.npz`
  - `grid_x`
  - `grid_y`
  - `posterior_z`
  - `pos_density_z`

## JSON metadata

- `params.json`
  - run configuration
- `metrics.json`
  - evaluation summary and metric metadata

## Python return value

When you use the library API rather than only the CLI, the workflow
returns typed public dataclasses from `psma.results`.