llms.txt
+16 −11
https://diditwork.org/llms.txt
− Site: https://diditwork.org · API: https://api.diditwork.org (read-only JSON, OpenAPI at https://api.diditwork.org/openapi.yaml) · MCP: https://api.diditwork.org/mcp (Streamable HTTP, JSON-RPC 2.0, 19 read-only tools, POST only). Product version 0.2.16.− - [Verdict procedure](https://diditwork.org/agents/verdict-procedure): how to answer "did it work?" about any giving claim, link by link (defined, happened, caused, sized, travels, worth it, adds, checked); report every link, let the weakest set the headline, never a score. JSON: https://diditwork.org/agents/verdict-procedure.json; donor version: https://diditwork.org/how-we-know#chain− - [What we cover](https://diditwork.org/agents/coverage.json): empty answers carry `coverage`; answers that cover only part of a question carry `partial_coverage` (which part is covered, which is not, who assesses the rest). We rate no humanitarian or disaster-relief charity and answer "is X a scam" only by pointing to the IRS, state regulators, BBB Wise Giving Alliance and the FTC.− Every donor page also links its machine-readable twin with `<link rel="alternate" type="application/json">`.− The detailed reference below covers the API, MCP tools, honesty rules and every donor page.− 13. **Link donors to a filtered list** instead of retyping one: `https://diditwork.org/charities/?cause=malaria&by=givewell` shows only matching charities, in the same order. Parameters: `cause` (a slug from https://diditwork.org/causes/), `by` (`givewell` or `hli`), `measure` (`lives`, `wellbeing`, `other`, `none`: what the headline figure measures), `followup=1` (only charities with a follow-up record), `q` (words in the name or description). Filters only hide; they never re-rank. The filter i− - MCP: https://api.diditwork.org/mcp (Streamable HTTP, JSON-RPC 2.0, POST only, stateless). **19** tools, v0.3.8: `list_charities` (the filtered charity list, same JSON as https://diditwork.org/charities/index.json), `get_meta`, `recommend_charity`, `list_leaderboards`, `get_review_queue`, `compare_charities`, `estimate_impact`, `get_org`, `search_orgs`, `get_trust`, `get_tax`, `allocate_portfolio`, `create_giving_intent`, `list_claims`, `get_claim`, `list_values_profiles`, `get_values_profile`,− - Method page: [How we know whether giving worked](https://diditwork.org/how-we-know) (JSON with formulas as LaTeX, definitions, norms for agents and sources: https://diditwork.org/agents/how-we-know.json). Part 1: counterfactuals, study designs, intervals and power, Bayesian shrinkage, cost-effectiveness uncertainty. Part 2: external validity and transport, attribution vs contribution (leverage, funging, Shapley), effects over time and discounting, Goodhart's law and survivorship bias, regressi− - Agent benchmark: https://diditwork.org/agents/benchmark (JSON: https://diditwork.org/agents/benchmark.json). 40 donor questions put to the MCP endpoint and scored against this site's own data (run run-20261025T181828Z): coverage 100.0%, correctness 97.4%, provenance 100.0%, honesty 100.0%, decision-readiness 98.2%; every failed check is listed with the fix queued. Alternatives are described from their docs and not tested.− - Question [Which charities do the most for people's wellbeing?](https://diditwork.org/questions/best-charity-for-wellbeing): Friendship Bench and StrongMinds are Happier Lives Institute Top Charities. Pure Earth and Taimaka are Happier Lives Institute Promising Charities. Friendship Bench 48.52 WELLBYs per $1,000 donated (Happier Lives Institute, published table; source https://www.happierlivesinstitute.org/friendship-bench-comprehensive-summary/; modeled, not guaranteed); StrongMinds 40.37 WEL− - Question [What's the best charity for mental health?](https://diditwork.org/questions/best-charity-for-mental-health): Friendship Bench and StrongMinds are Happier Lives Institute Top Charities. Friendship Bench 48.52 WELLBYs per $1,000 donated (Happier Lives Institute, published table; source https://www.happierlivesinstitute.org/friendship-bench-comprehensive-summary/; modeled, not guaranteed); StrongMinds 40.37 WELLBYs per $1,000 donated (Happier Lives Institute, published table; source htt+ Site: https://diditwork.org · API: https://api.diditwork.org (read-only JSON, OpenAPI at https://api.diditwork.org/openapi.yaml) · MCP: https://api.diditwork.org/mcp (Streamable HTTP, JSON-RPC 2.0, 22 read-only tools, POST only). Product version 0.2.16.+ - [Verdict procedure](https://diditwork.org/agents/verdict-procedure): how to answer "did it work?" about any giving claim, link by link (L0–L7); report every link, let the weakest set the headline, never a score. JSON: https://diditwork.org/agents/verdict-procedure.json; donor version: https://diditwork.org/how-we-know#chain+ - [Evidence gaps](https://diditwork.org/agents/evidence-gaps.json): for each rated charity and each link of the verdict chain, what our cited sources show, what perfect information would show, the gap (ordinal words, never a score) and whether closing it could change the donor's decision. Also on `get_org` as `evidence_gaps`. Explainer: https://diditwork.org/how-we-know#ceiling+ - [Evaluator status check](https://diditwork.org/agents/evaluator-status.json): when we last re-read each followed evaluator's own list of recommended charities and its news (GiveWell, Happier Lives Institute), per evaluator `last_verified_at` and `result` (match / mismatch / fetch_failed), open corrections and the status notes we show (e.g. a paused recommendation, in the evaluator's words). Same object as `evaluator_status_check` on `GET /v1/meta` and MCP `get_meta`. If it is old, check the ev+ - [What we cover](https://diditwork.org/agents/coverage.json): empty answers carry `coverage`; answers that cover only part of a question carry `partial_coverage` (what is and isn't covered, who assesses the rest). Search answers (`search_orgs`, `GET /v1/search`) that don't fully cover the question start with `answer_first`: say it first. Asked who funds us, they also carry `funding_question`: nobody can pay for a place, no referral fees, we never handle donors' money; who pays to run the site i+ - [One-call decision brief](https://api.diditwork.org/v1/brief?question=malaria): MCP `decision_brief` = `GET /v1/brief`. Say `answer` first; shortlist in site order, never our ranking. "claim real?" gets `how_to_judge`; "vouch for X" `vouch_request`; "one study beats the rest?" `star_result`; "rank X first/guarantee/rate 10/10" `request_declined`; "average the costs per life" `average_declined`; "give now or wait?" `give_now_or_wait`; "a match?" `marginal_dollar`; "fund or pick myself?" `fund_o+ Every donor page links its JSON twin via `<link rel="alternate" type="application/json">`.+ Below: the API, MCP tools, honesty rules and every donor page.+ 13. **Link donors to a filtered list**: `https://diditwork.org/charities/?cause=malaria&by=givewell` shows only matching charities, same order. Parameters: `cause` (a slug from https://diditwork.org/causes/), `by` (`givewell` or `hli`), `measure` (`lives`, `wellbeing`, `other`, `none`: what the headline figure measures), `followup=1` (only charities with a follow-up record), `q` (words in the name or description). Filters only hide; they never re-rank. Empty, unknown or duplicate values are neve+ - MCP: https://api.diditwork.org/mcp (Streamable HTTP, JSON-RPC 2.0, POST only, stateless). **22** tools, v0.3.9: `verdict_chain` (one charity's chain L0–L7, copied from our data, no computed figure or overall verdict; `GET /v1/verdict-chain/<org_id>`), `decision_brief` (one call, one defensible answer: `answer`, cited shortlist, evaluator fund, `units_differ`, `quote`), `list_charities` (the filtered charity list, same JSON as https://diditwork.org/charities/index.json), `get_meta`, `recommend_+ - Method page: [How we know whether giving worked](https://diditwork.org/how-we-know) (JSON with formulas as LaTeX, definitions, norms for agents and sources: https://diditwork.org/agents/how-we-know.json). Part 1: counterfactuals, study designs, intervals and power, Bayesian shrinkage, cost-effectiveness uncertainty. Part 2: external validity and transport, attribution vs contribution (leverage, funging, Shapley), effects over time and discounting, Goodhart's law and survivorship bias, regressi+ - Chain card: [The chain card](https://diditwork.org/chain-card) (JSON: https://diditwork.org/agents/chain-card.json). The 8 links of the verdict chain on one printable page, each with its question, what to ask the charity and the rule it carries, copied verbatim from https://diditwork.org/agents/verdict-procedure.json. A method, not a score.+ - Cause map: [Survive and thrive: a map of where giving can help](https://diditwork.org/survive-and-thrive) (JSON: https://diditwork.org/agents/survive-and-thrive.json). A mutually exclusive, collectively exhaustive map: Survive = deaths by IHME GBD 2023 cause group (Levels 1-2); Thrive = six domains of life among the living; future catastrophic risks on a separate axis; animals. Per group: health aid tracked by IHME (or not_tracked / not_yet_sourced), aid-share-to-death-share ratios with their + - Agent benchmark: https://diditwork.org/agents/benchmark (JSON: https://diditwork.org/agents/benchmark.json). 40 donor questions put to the MCP endpoint and scored against this site's own data (run run-20261027T220239Z): coverage 100.0%, correctness 97.4%, provenance 100.0%, honesty 100.0%, decision-readiness 100.0%; every failed check is listed with the fix queued. Alternatives are described from their docs and not tested. Verdict-chain items (https://diditwork.org/agents/benchmark#chain, JSON+ - Question [Which charities do the most for people's wellbeing?](https://diditwork.org/questions/best-charity-for-wellbeing): Friendship Bench and StrongMinds are Happier Lives Institute Top Charities. Pure Earth and Taimaka are Happier Lives Institute Promising Charities. The Happier Lives Institute has paused its recommendation of Friendship Bench while it reassesses. Friendship Bench 48.52 WELLBYs per $1,000 donated (Happier Lives Institute, published table; source https://www.happierlivesins+ - Question [What's the best charity for mental health?](https://diditwork.org/questions/best-charity-for-mental-health): Friendship Bench and StrongMinds are Happier Lives Institute Top Charities. The Happier Lives Institute has paused its recommendation of Friendship Bench while it reassesses. Friendship Bench 48.52 WELLBYs per $1,000 donated (Happier Lives Institute, published table; source https://www.happierlivesinstitute.org/friendship-bench-comprehensive-summary/; modeled, not guaranteed);