Guide to the sections
What each part of the site is for, and when Evidence, Explore or a synthesis is the right place to look.
Several pages overlap. This is the short version of which one to open for which kind of question.
Evidence or Explore
These two are the most commonly confused, and the distinction is just the unit of analysis.
- Evidence — one claim at a time
- Every extracted finding, filterable by direction, confidence and theme. Use it for a specific question: "has anyone found anything about AI and wages?" Each claim links back to its paper.
- Explore — one category at a time
- Whole outcome categories compared against each other, sortable by how much the evidence agrees, how strong it is, how fast it is growing and how contested it is. Use it for the lay of the land rather than a single finding.
Everything else
- Papers
-
The whole corpus, newest first. Filter by source, evidence strength and study
design, or search it. Each paper opens to its assessment, claims and any
linked datasets.
Search ranks semantically, so a question phrased in your own words reaches papers that share none of its wording. The trade is that a query always returns its nearest neighbours rather than an empty result, so a full page is not evidence that anything on it answers your question. See the caveat. - Syntheses
- Written state-of-the-evidence pages for categories with enough behind them. A synthesis reads the claims for you and reports where they agree, where they conflict and what is missing. The best starting point for a topic you do not know.
- Digests
- The weekly write-up of what arrived and why it mattered. See the digest labels for how to read the tags on each entry.
- Trends
- Aggregate views over the corpus: methods, coverage, impact and authors. See trends and analytics for what these can and cannot tell you.
- References
- Check a bibliography against an economics corpus to catch fabricated or retracted citations. See the reference checker.
Author and collaboration views
Author pages and the co-authorship network are built from corpus records, not from verified identities. Exact provider IDs and unambiguous ORCIDs support conservative links; names alone never merge identities, so two researchers who share a name stay separate. Provider coverage differs, so paper and collaborator counts change when you change the provider selector. Read those views as "what this corpus records", not as a biography.