Methodology

How a click on the map becomes one concrete next step — with sources, schedules, and the audit trail anyone can inspect.

Data sources

Every topic uses only freely accessible, primary public data. We don’t scrape, crawl, or bypass rate limits. Per-topic citations live on each topic’s About page; the recurring sources are:

Daily refresh cadence

Each topic has a small scheduled AWS Lambda that pulls its data once a day at the source’s published cadence. The Lambdas are staggered through the day so we never burst on any single upstream provider:

  • 06:33 UTC — air
  • 06:47 UTC — bees
  • 10:17 UTC — fire
  • 14:15 UTC — forest
  • 16:41 UTC — food
  • 19:31 UTC — ocean
  • 23:32 UTC — water

Public APIs are hit at most once per day no matter how many people are reading the map. We are deliberately a quiet, welcome client of the data sources we depend on.

Donations and third-party organisations

When a map suggests donate or advocate, the goal is to turn public data into one honest next step — not to rank charities or collect money on anyone’s behalf.

  • Human-authored catalogs. Outbound organisations live in per-topic actions.json files in the repository. Maintainers add only well-known public NGOs, UN agencies, government programmes, or open communities with published accounts and a clear mission fit. We do not accept affiliate links, paid placement, or product pitches.
  • Rule-based picks, not chatbot prescriptions. Country cards combine live open-data signals (e.g. IFRC GO active operations, WHO/UNICEF JMP gaps, WFP hunger classifications) with simple thresholds written in code. The highest-priority signal wins; there is no LLM choosing where your money should go at click time.
  • Daily link verification. The same scheduled Lambda that refreshes indicators also HTTP-probes every catalog URL. Working links are marked verified today; inconclusive probes stay visible with an unverified today badge; confirmed-dead links are dropped from the published catalog. The full audit is public at <topic>.actsmall.org/live/curator.json.
  • No endorsement. Listing an organisation means editors believe it is a reasonable, transparent place to start — not that ActSmall guarantees outcomes, tax treatment, or fitness for your situation. You are responsible for your own due diligence.
  • Emergency vs chronic giving. When IFRC GO reports a live operation in a country, we surface that operation first (API-refreshed daily). Otherwise chronic gaps may point at long-running WASH or hunger operators from the curated catalog. National statistics can miss local realities; verify with trusted local sources when you can.

Language-model tools may help maintainers draft site copy under human review. They are not used to generate donation targets or to rewrite the action catalogs without an editor.

The recommender

The recommender is deliberately simple. From a tap location it derives need-signals (topic-specific: drinking-water gap, PM2.5 ratio, IFL loss, hunger gap, beehive density, etc.), and ranks the per-topic catalog of citizen actions by use-case match plus simplicity. Lower cost and lower required skill rank higher. It is a discovery tool, not a prescription.

National figures hide local realities. We say so on every country card. For the United States, drill down with the relevant federal data dashboard (EPA ECHO, USGS NWIS, etc.). For other countries, ask local authorities and use the per-topic country snapshots in the library where available.

Link verification

The same daily Lambda that pulls public data also probes every external URL across the topic’s editorial catalog and library. Outcomes per link, written to a public audit log:

  • OK — HTTP probe confirmed the URL resolves; kept as-is.
  • Repaired (apex-walk) — the original URL 404’d but the same path under the apex domain resolves; we redirect with a note in the audit log.
  • Repaired (Wikidata P856) — the original URL is dead but Wikidata has an official-website property for the same entity; we use that.
  • Repaired (AI suggestion, verified) — both of the above failed, so we asked AWS Bedrock Nova Micro for a candidate replacement. The candidate must independently pass an HTTP probe and a semantic plausibility check (registrable domain shares a meaningful token with the original label, host, or organisation name). We have never accepted an unverified AI suggestion.
  • Unverified today — the probe was inconclusive because the source soft-blocked our checker behind a bot wall. We keep the maintainer’s vetted link but flag it in the audit log so anyone can re-check.
  • Dropped — the URL is dead and no plausible replacement exists.

The full audit log of every URL checked, with status and any repair, is published at <topic>.actsmall.org/live/curator.json for anyone to inspect.

AI tooling — disclosure

The written text on this site is authored by humans and reviewed by humans. We may use language-model tools (Anthropic Claude, OpenAI, AWS Bedrock) to draft outlines, suggest rewrites, generate alt-text, and explain jargon — all then reviewed and edited by a human before publication. The text is the source of truth.

The shared link curator described above does call AWS Bedrock Nova Micro as a last-resort replacement suggester for confirmed-404 URLs — but every AI-proposed replacement must independently pass an HTTP probe and a semantic plausibility check before it’s accepted. The check rejects more candidates than it accepts.

We do not run LLMs at request time on visitor input. The site has no chat, no AI assistant, no “ask me anything” input box. All AI use happens once a day in the maintenance pipeline.

No-data handling

When the country you click has no current reading and no editor-vetted national-context citation, we don’t guess and we don’t leave the panel blank. We surface a quiet, neutral “help open the data” callout pointing at the relevant open-data effort — OpenStreetMap tagging, iNaturalist / GBIF observations, the Internet of Water Coalition, citizen-science networks. The absence of data is itself an action one click can chip away at.

See /contribute/ for the cross-topic ways to help open the data.

If you find an error

If you find an error, an out-of-date source, a broken link, or a claim that overstates the underlying evidence: please email submissions@actsmall.org with the page URL and what looks wrong. The maintainers will review.