Technical reference. Written for analysts and AI agents and kept in sync with the repository, so it uses internal terms. Raw Markdown: /pages/vision.md.

Vision: "Did it work?"

Better World (diditwork.org): where we are going, plainly stated. Last updated 2026-10-21.

The problem

People, and more and more often the AI assistants acting for them, ask "where should I give?" They usually get a confident answer with no source, no date, and no way to check later whether the money did what was promised.

What we want to be by 2029

The place that donors and their agents check first, because every answer here is sourced, dated, labeled, and re-checked, and because nobody can pay to change it.

How we plan to get there

  1. Keep a public "Did it work?" ledger. When a recognized evaluator (GiveWell, the Happier Lives Institute, Animal Charity Evaluators, WHO, IHME) publishes an estimate, we record exactly what they said and when, with a hash so the record cannot be quietly edited, and a date to check again. Over the years this becomes a history of which promises held up. Today it holds 9 seeded records, all copied from evaluator citations we already had. None of them has been re-checked yet.
  2. Publish an open format for impact claims. It is a simple, free standard (license pending) plus a checker that any charity can run on its own claims: number, unit, source, date, uncertainty, and when to re-check. Passing the checker means a claim is well-formed and honestly labeled. It does not mean the claim is true.
  3. Be the tool AI assistants call. We publish machine-readable interfaces (an API, MCP tool descriptions, llms.txt) so assistants can quote us accurately instead of guessing.
  4. Learn what donors want, privately. We have proposed (not switched on) counting only coarse, anonymous totals, such as how often a cause is asked about, with small groups hidden and an easy opt-out, so charities and evaluators can see unmet demand.

What we will not do

  • We will not invent cost-effectiveness numbers. When a number isn't published by a recognized evaluator, we say "unknown."
  • We will not present a modeled estimate as a promise. Modeled, not guaranteed.
  • We will not let money buy rank. Sponsored or paid placement can never change rankings or evidence.
  • We will not combine different kinds of good (lives saved, wellbeing, carbon) into one score.
  • We will not move donors' money, take charges, or give tax advice.

Where things stand today (honest)

  • The site and the read-only API are live (https://api.diditwork.org). The ledger is new and small.
  • Ranking covers 50 global-health organizations, using GiveWell and Happier Lives Institute evidence.
  • Charities cannot submit claims yet. That waits on an approved review process.
  • One person (Ho Hum) approves every change to which organizations are listed or ranked.

How to hold us to this

Every record and claim links to its source. The ledger's hash chain shows whether history was edited. If you find an error, tell us: errors count against us, and a public corrections log is planned.