Search the web for surveys to validate an Audience against (experimental)

Experimental. Needs an included validation left this month, or synthetic responses for the smallest validation run (8 questions for every ready Mind). Starts a background search for published survey results that match the Audience: queries are planned from its name, brief and member composition, the web and the public source bank are searched, and each source is read once. A source becomes a benchmark only when every kept share is backed by a verbatim passage of the source, the question wording appears in it, single-choice shares add up, and a base is stated. Results are cached per source text and shared with later searches from any Audience. Returns 202 with the search; poll GET /audiences/validation-benchmarks?audienceId= (its summary.search) and the found surveys appear as kind: public candidates with discovered: ids. The search also judges, for every survey on offer, who it asked and whether its questions bear on the Audience. A search already running, or completed within the last day with every offered survey judged, is returned instead of starting another unless force is true.