Evidence BenchmarksSubmit a platform

Open benchmarks for evidence synthesis

How accurately do platforms find, read and summarise development research? Each benchmark scores them against what published reviews concluded.

Finding agreement

Theory of change · Higher is better

No platform scored yet

Be the first →

Links drawn

Theory of change · Higher is better

No platform scored yet

Be the first →

Steps drawn

Theory of change · Higher is better

No platform scored yet

Be the first →
Open to allPut your platform on the leaderboardAnswer the questions, run the public scorer, open a pull request. We publish what you submit.Gold from published reviewsEvery gold value is quoted, with its page, from a published review or dataset. Nobody labels by hand.

Changelog

  1. New benchmark · Nov 20Theory of change

Theory of change

Gold and scorer ↗

Does the platform draw the steps a published review's theory of change has, and does it judge the evidence on each link the way the review does?

3 questionsPublished 3ie and Campbell systematic reviews that test the evidence for each link of the intervention's causal chaingold ready, scorer to build

Finding agreement

Of the gold links the review assessed, the share where the platform draws the link and its finding (benefit, no clear effect, harm, mixed) matches the review's. · Higher is better

No platform has been scored yet

The questions are ready. How to submit a platform.

Method

Gold from published work

Every gold value is what a published review, dataset or guideline says, with a verbatim quote and its page. Where the source says nothing, the gold says nothing.

The same questions and scorer

Every platform answers the same questions in the same format, and scores them with the same public scorer.

Trusted, self-run submissions

Each platform runs the benchmark itself and submits its answers and scores. † marks a run whose platform had seen the gold or its sources before answering.

Built 2026-11-20.