The Clearinghouse collects recommendations produced by citizens' assemblies,
panels, and other deliberative mini-publics — groups of randomly selected citizens
who deliberate on a policy question and issue formal recommendations. Its purpose
is to make that body of citizen-authored knowledge searchable and citable by the
people who can act on it: public authorities, NGOs, researchers, and civil society
organizations.
Trust is the product. Every recommendation shown here is a
verbatim quote copied directly from the assembly's own final report, anchored to
the exact page where it appears. We do not paraphrase, translate, or improve the
wording — if a recommendation reads awkwardly in the original, it reads awkwardly
here too, because rewording it would be a small act of interpretation, and
interpretation is exactly what we're not qualified to do on citizens' behalf.
Extraction. Recommendations are pulled from source PDFs by a
language model instructed never to invent, estimate, or summarize the operative
text — only to locate it. Every extracted quote is then automatically checked
against the PDF's own text layer: if the quote can't be found on the page it claims
to come from, it is flagged rather than silently trusted. You may still see
recommendations marked "needs review" — meaning a human has not yet manually
verified the automated extraction. Treat the linked source PDF as the final
authority.
Classification. Fields like policy domain, target actor, and
geographic scope are the model's best judgment, meant to help you filter and
browse — treat them as a rough index, not certified metadata the way the verbatim
text is.
Support data (vote shares, consensus statements) is recorded
only when the source report explicitly states it. We never estimate or infer a
figure — no support data shown means the report itself didn't quantify it.
Sources. The corpus currently draws on the OECD's database of
deliberative processes, direct submissions from assembly organizers, the Missions
Publiques archive, and reports contributed through this site. We link out to
primary sources rather than duplicate process-level data already maintained
elsewhere (OECD, Participedia, and others) — the Clearinghouse is a recommendations
layer, not a competing process database.
What this is not: a ranking, an endorsement, or a claim that any
recommendation is more "correct" than another. It surfaces what citizens said, with
enough provenance that you can judge for yourself whether it's relevant to your
work and cite it with confidence.
Corrections. If you spot a misquote, a wrong page reference, or
a document that shouldn't be here, use the Contribute flow to flag it — every
correction makes the corpus more trustworthy for the next person who uses it.