A research lab by EasyByte
Measured,
not assumed.
Open studies and verification tools for the rules software has to follow.
Latest
New findings
All 11 studies →
S17 · High-value datasets
7 of the 27
Member States for which the European data portal identifies no high-value dataset
High-value datasets on data.europa.eu: what a machine sees, and what Member States say
S11 · Tourist-licence numbers
6.6%
of Airbnb listings in five Spanish areas with a well-formed tourist-dwelling number show one we could not trace to any registered dwelling
Do the registration numbers on Airbnb listings exist in the official registries? Spain and New York around 20 May 2026
S10 · NeTEx timetables
8 of 34
timetable datasets from five European national access points are fully valid against the NeTEx XSD, all of them French
Past the static deadlines: NeTEx on five European access points, checked against open schemas
Findings
Every headline figure, with its sample
| Figure | Finding | n | Study |
|---|---|---|---|
| 7 of the 27 | Member States for which the European data portal identifies no high-value dataset | 27 Member States, 27,972 HVD records | S17 |
| 6.6% | of Airbnb listings in five Spanish areas with a well-formed tourist-dwelling number show one we could not trace to any registered dwelling | 2,066 of 31,222 | S11 |
| 8 of 34 | timetable datasets from five European national access points are fully valid against the NeTEx XSD, all of them French | 34 datasets, 5 access points | S10 |
| 32.9% | of Spain's public-sector home pages on HTTPS send HSTS | 1,688 of 5,130 | S8 |
| 5.0% | of Google AI Overview answers on recently changed Spanish rules were outdated or wrong | 12 of 240 | S7 |
| 50.2% | of AI images uploaded to Wikimedia Commons in June–July 2026 carry a machine-readable AI mark | 313 uploads | S6 |
| 23.2% | of the top PyPI projects attest their latest version | 3,479 of 14,995 | S5 |
| ≈0.9% | of Wikidata's multilingual drug labels are wrong, a lower bound | 34,207 labels | S4 |
| 124/124 | pixel or container rewrites removed the embedded C2PA manifest | 124 rewrites | S3 |
| 39.3% | of requests claiming to be an AI or search crawler were spoofed | 16,548 claims, 3 sites | S2 |
| 28.6% | of the files presented as valid Verifactu records contain an error | 35 files | S1 |
Topics
Browse by topic
Tools
A verdict that cites the rule
All tools →ai-mark-lintChecks that synthetic images, video and audio still carry the machine-readable AI marks the law asks for, and cites the article for every finding.attest-lintWarns when a pinned Python dependency stopped shipping PEP 740 attestations on PyPI, or when its Trusted Publisher changed.crs-lintLints OECD Common Reporting Standard XML offline against the official XSD and the Status Message business rules, citing the official OECD error code.