govpl: drop the prosecutor offices' press releases
Drops the press releases of gov.pl's 45 prosecutor-office sections (Prokuratura Krajowa, the pr-* regional and po-* district offices) from govpl.
Why: these releases report charges, arrests, indictments and sentences of people who can often be identified, even when a release gives only a first name and a surname initial. That is personal data about offences (GDPR Art. 10). In these sections 62.9% of rows match a criminal-procedure term regex; in the rest of govpl 1.4% do.
Counts: 87,816 → 74,946 documents, 216,869,817 → 185,107,788 characters, 80,204,973 → 68,291,645 tokens (−11,913,328).
How:
fetch_govpl.PROSECUTOR_SECTIONis the one rule.fetch_govpl.pyno longer fetches these sections.clean_govpl.pydrops their rows after the PAP rule and the third-party review, from the same pinned inputs, and records them as the newdrop_prosecutorstat.- Ids are unchanged. The new Parquet and sidecar equal main's minus exactly these 12,870 rows; this was checked row by row. No text was fetched from both a prosecutor section and another section.
- 3 rows that the third-party review had kept are among them.
Updated: the card; the src/sources.py entry, including the footer-period counts, which were recomputed as 35,207 / 24,201 / 7,381; artifacts/source_findings.md; and the contract tests (new rule test, drop_prosecutor checks, no prosecutor subsite in the sidecar).
Not covered: in other sections, 20 rows match both the criminal-term regex and the initials pattern. They have not been reviewed.
🤖 Generated with Claude Code