AIUS can inspect CSV, TSV, Parquet, JSON and JSONL files, profile quality,
query declared local files with DuckDB, compare datasets, audit train/test
splits, and examine drift. It can also create, edit, execute and export
Jupyter notebooks when the required local Python packages are installed.Training and validation tools record dataset hashes, package versions, fold
membership, metrics and model hashes. Those records help you review an
analysis; a passing split audit or cross-validation score is not proof of
causal effect or deployment performance.The terminal’s scientific tools run locally with your permissions. Review
file and execution requests, especially for untrusted notebooks or saved
models. Uploading an output to the web platform requires separate approval.
