DataSentinel

Data quality and ML monitoring for a daily batch of loan applications, characterised by the smallest fault it can actually see rather than by how often it catches an obvious one.

Loading.

What this is. Synthetic data throughout. The faults were injected by the author and the thresholds set by the author, so this is a real test of the wiring and a weak test of the world. The benign scenarios and the false-alarm counts are what keep it from being circular; the detection floors are what make it informative.

GitHub Pages serves files rather than processes, so this page reads a committed snapshot of one pipeline run. It is a record, not a live console.

Source and measurement methodology on GitHub