An open dataset of the EU trusted lists, as 5 instruments saw them

What is in it

• Observations: 14,539 rows, one per attempt to fetch one document, with the outcome, the error class, the size and the SHA-256 digest of what came back.
• Documents: 896 distinct versions (856 XML lists, 39 PDF renderings and 1 HTML page), with header fields and provider and service counts read from the bytes, and a public copy of every XML version.
• Derived tables, code and a methods note: the script that rebuilds the derived tables from the package, and its result.

The five instruments are Cassandra (a daily fetch from a GitHub-hosted runner, 141 runs between 20 May and 19 October 2026); the RQSCD and QEAA censuses, run from a server the Tyche Institute operates; TLHealth, which checks HTTP and TLS reachability of every pointer in the EU list; and Wayback Machine captures of every list address the EU list has announced, 2010–2026.

What the methods note shows

On 27 days the published daily total of trust service providers fell 5 or more below its median, by as much as 96 on 12 June 2026. None of the 27 is a removal. Each time, one or more lists did not download. On all 141 runs, the published totals equal a recount of the stored bytes by separately written code. Adding back each failed list's last good copy brings the total back to the median exactly on 24 days; on the other 3 the underlying count rose that same day. A day-to-day difference of fetched lists therefore cannot, on its own, tell a provider that left from a list that did not load.

Personal data

Trusted lists embed certificates of natural persons. In the public copy, certificates, subject and issuer names, national identifiers, names written as distinguished names and e-mail addresses are replaced by deterministic tokens. Certificate tokens follow the rule of 10.5281/zenodo.22389154; every other token is keyed, so that names and identifiers cannot be guessed back. The 39 PDF renderings are not redistributed. Provider names and addresses stay as the Member States publish them, because they are the content of the public register.

Use of AI tools

AI coding assistants (Claude, Anthropic) helped write the instruments' code and the scripts and tables of this package, under the author's direction. The author defined the study, reviewed the results and is responsible for the dataset.

Citation

Sokolov, A. (2026). EU trusted lists observed by five instruments, 2010–2026 (Version 1.0.1) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.23473890

@dataset{sokolov2026eutrustedlists,
  author    = {Sokolov, Anton},
  title     = {{EU} trusted lists observed by five instruments, 2010--2026},
  year      = {2026},
  month     = oct,
  version   = {1.0.1},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.23473890},
  url       = {https://doi.org/10.5281/zenodo.23473890}
}