Grantmaking across public corpora

Research output, public funding and policy, discourse, forecasting, and public identity are queryable together, each row carrying its source identity and provenance.

The live schema (GET /v1/scry/schema) is the exact, current list; a source absent there is not served.

One source answers a narrow question. The decisive move is the join: connect a funded project to its publications, the publications to their authors, the authors to what they argue in public, and that to what forecasters expect.

• Funding to output Join grant records to OpenAlex works and full-text papers to see what funding actually produced.
• Output to people Resolve paper and post authors to cross-platform public identities.
• People to discourse Read what those people argue in public on forums and newsletters.
• Signal to timing Weight by recency and by what Polymarket, Kalshi, and Manifold imply about the field.

A name on a paper, a forum handle, and an X account are often one person. Scry clusters public accounts into persons only on strong shared identity evidence, so a shortlist is people, not strings. Common names are never silently merged.

• Deterministic only Accounts cluster only on strong shared identity keys, such as self-declared cross-account links and canonical IDs.
• Evidence named Each cluster states the kinds of evidence behind it, never the key values themselves.
• Person surfaces persons.links serves linked public accounts to approved enterprise accounts ([email protected]); the persons.link_coverage aggregates are open to everyone. Accounts are joined when they publish the same identity key — an ORCID, a homepage, an explicit cross-link — or, for a few platform pairs, carry the same handle where that equality was measured at least 93% precise (marked handle_lex in the evidence kinds). No stylometry, no text inference.

A candidate list is the start. Rerank it against the criteria that matter for a decision, using calibrated pairwise judgment rather than lossy one-to-five scores. Reviewed judgments carry confidence, uncertainty, and links back to the evidence they rest on.

This builds a candidate list for existential-risk philanthropy. It seeds from direct-philanthropy terms, existential-risk terms, and counterfactual-leverage terms across public forum discourse, then ranks authors by evidence volume, topic breadth, and recency. Every row carries the URIs it came from.

WITH seed_hits AS (
  SELECT 'direct_philanthropy' AS seed, 5.0 AS seed_weight,
         source, original_author, title, uri, original_timestamp
  FROM forums.posts
  WHERE original_author IS NOT NULL AND original_author != ''
    AND multiSearchAnyCaseInsensitive(payload,
        ['grantmaking', 'open philanthropy', 'philanthropy'])

  UNION ALL

  SELECT 'existential_risk' AS seed, 4.0 AS seed_weight,
         source, original_author, title, uri, original_timestamp
  FROM forums.posts
  WHERE original_author IS NOT NULL AND original_author != ''
    AND multiSearchAnyCaseInsensitive(payload,
        ['existential risk', 'x-risk', 'ai safety', 'biosecurity'])

  UNION ALL

  SELECT 'counterfactual_leverage' AS seed, 3.0 AS seed_weight,
         source, original_author, title, uri, original_timestamp
  FROM forums.posts
  WHERE original_author IS NOT NULL AND original_author != ''
    AND multiSearchAnyCaseInsensitive(payload,
        ['forecasting', 'neglectedness', 'field building', 'public goods'])
)
SELECT
  original_author,
  count() AS evidence_rows,
  uniqExact(seed) AS topic_breadth,
  groupUniqArray(seed) AS matched_lenses,
  groupUniqArray(source) AS sources,
  max(original_timestamp) AS latest_seen,
  round(sum(seed_weight) + 3.0 * uniqExact(seed), 2) AS candidate_score,
  arraySlice(groupUniqArray(uri), 1, 5) AS evidence_uris
FROM seed_hits
GROUP BY original_author
ORDER BY candidate_score DESC, latest_seen DESC
LIMIT 50