Show HN: Ato – A thin ML layer. When results change, it tells you why

Hi HN. I built Ato because I wanted a layer that works wherever I attach it, and tells me why things changed when they do.

I'm bad at rituals (perfect logging, tidy checkpoints). Ato still gives me answers by fingerprinting three pieces:

- Config structure: architecture-level changes, not just values - Code logic: function-bytecode fingerprints, not commit IDs - Runtime outputs: detects silent drift even when code "didn't change"

No dashboards, no servers; just SQLite and a CLI. It sits beside Hydra/argparse/MLflow/W&B rather than replacing them.

If your runs sometimes diverge or fail and you just want "why" without adopting a platform, I'd love your feedback.

Thanks for checking this out!