data/methodology.md

# Methodology

The release starts from public-source article discovery and a stricter sensitive-site event view. Rows are ranked into evidence tiers rather than flattened into one confidence level.

The public table removes local file paths, runtime details, and internal run metadata. It keeps only source-facing fields: headline, publisher, URL, report date, site signal, country/region, evidence tier, cluster id, and a row hash.

Evidence tiers:

- `resolved_sensitive_site_report`: the stricter event view matched the source to a sensitive site candidate.
- `named_sensitive_site_report`: the source names a sensitive site, but the row is still a candidate that needs follow-up.
- `source_discovered_report`: public source title or URL language indicates a relevant drone report around a sensitive context; use as a lead for review.