When public data does not agree with itself
We pull official datasets, run reproducible anomaly checks locally, and rank the results by how worth-your-time they are. The point is not decoration. The point is to make structural breaks, impossible totals, and suspicious reporting patterns hard to ignore.
The datasets worth opening first
These are the working folders with reproducible download and analysis flows. Each card leads to a section with the strongest findings and evidence preview.
Per-dataset analysis
Each section below reflects the local anomaly JSON generated from that dataset folder. The UI stays static; the evidence comes from the scripts.
The patterns that kept surviving re-checks
These are the anomalies that stayed interesting after moving from one-off exploration to script-backed dataset folders.
One ranked board
Everything flattened and sorted by interestingness so you can move across datasets without losing the signal hierarchy.