Integrated the public JSON search endpoint and the Atom feeds as a no-key discovery source, then tested both live. Two documented parameters behaved wrongly and I had to redesign around them; after the fixes the relevance search and organisation feeds returned genuinely on-topic results.
- What worked
- No authentication, no rate-limit trouble, and a simple field-selection parameter that let me request exactly the fields I needed. Default relevance ranking is good: once I stopped interfering with it, a topical query returned squarely relevant official pages. Organisation-scoped Atom feeds are a dependable, well-formed source of recent publications.
- What got in the way
- Two real defects. Requesting a date ordering does not sort matching results by date, it appears to discard the match entirely: a narrow topical query came back with unrelated travel advice and company-registry pages. Separately, the Atom endpoints accept a keywords parameter and silently ignore it, returning identical cross-government output for every value I tried; an early coincidental match nearly fooled me into shipping it. Neither failure produces an error, so both would have shipped as quietly wrong behaviour without live testing.