Three public datasets hard to reach by machine, Oct 2026
Three state datasets, three different barriers to automated civic access
Three public datasets that were hard to reach by automated request (October 2026)
Three datasets that planning evidence depends on are public in law and hard to reach by machine, each in a different way.
| Dataset | What happened (probes of 4 October 2026; the browser visit that day has no recorded time) | Form of barrier |
|---|---|---|
| HM Land Registry INSPIRE title boundaries | Automated requests got 403 at a Cloudflare edge in 25–87 ms. A browser request reached the origin, which answered “Service WFS is disabled”. | Edge block for scripted requests |
| HSE Major Accident Hazard pipelines | The pipeline found no national programmatic service; it used two council layers. HSE’s own advice tool is web-based and open to planning authorities and developers, so what is missing is a machine-readable national feed, which cannot be proved by probing. | Fragmented local provision |
| Coal and mining risk (Mining Remediation Authority) | The public catalogue answers, but 3 of 4 configured service URLs return an error message (HTTP 200 with an error body) and the fourth returns no feature at the test point. A layer found through the catalogue does work. | Endpoint instability |
The same fragility in our own data
The flood zone could not be determined for 574 of 652 sites (88.0%). The Environment Agency endpoints answered with an error message inside an HTTP 200 response, which the code read as an empty result, and the fallback did not return a usable zone. A later re-query on the planning platform determined every site (600 with no flood zone 2 or 3 polygon at the point, 52 with one). We removed flood from the analysis rather than read “unknown” as “safe”.
What we can and can’t say
The Land Registry response states no reason. The commercial-licensing explanation (Ordnance Survey MasterMap) rests on documentary sources, not on the response. What we document directly is the effect: ownership data that is open in law is out of reach of an automated query inside a 21-day objection window. Our ownership analysis used bulk releases instead, a workaround a resourced pipeline has and a community doesn’t.
We also corrected ourselves in public. An earlier draft reported a 100% Coal Authority failure and attributed it to revenue protection. A live probe overturned that: the catalogue publishes a queryable layer, and the pipeline obtained 28 positive hits from it. The claim was withdrawn and restated as endpoint instability. A separate NaPTAN timeout figure had no traceable source and was dropped.
Revised 12 November 2026: Title changed to say what was observed and when. Wording otherwise unchanged.
Evidence type: Direct count. Source: Manuscript ch. 6.3, 7.1. No claim in this series is causal.