Methods that can be inspected, rerun and challenged.

Preregister before modelling

Confirmatory hypotheses, outcomes, models and expected directions are written down before analysis. Any later deviation is reported with the paper.

Pin every input

Public-data snapshots are recorded by source URL and checksum. The committed manifest is the receipt; raw data remains outside Git.

Generate the published artifacts

The Python workspace produces versioned JSON artifacts. The website validates those artifacts before rendering; published prose never reaches directly into working data.

Report nulls and limitations

Results that do not support a hypothesis remain part of the record. Observational associations are not presented as causal effects.