Practical SolutionsEvidence for what works

Method

This page describes exactly how the site is made. Nothing here is hidden behind a model: where a model is used it is named, and today the loop uses none.

1. How problems are chosen

A problem is on the list when a public global indicator (World Bank open data, world aggregate) sits away from an international target (SDG, WHO, World Bank). The score is the shortfall to target, between 0 and 1; the ranking is computed by code on every refresh, not edited by hand. The list of indicators is code in problems.js; adding one is a pull request. Limits: it ranks gap, not deaths or years of life lost, and ignores how solvable a problem is. A problem whose indicator has no usable data is still shown, marked as unranked.

2. How solutions get in

An entry is added only with a citation to a study, the study's own figures, where it was tested, what it takes, who can act, a cheapest first test, and its limits. Entries are written in code, so every addition is reviewed and recorded in version history. The automated loop proposes leads; it does not promote a lead to an entry on its own.

3. Labels and evidence grading

  • Systematic review or meta-analysis: Pools many studies; the strongest design cited.
  • Randomised trial: Participants or places assigned by chance.
  • Observational study: Compares groups without random assignment; confounding is possible.
  • Expert guidance: A recommendation, not a measurement.

How to read the labels

  • evidence-backed Cites a systematic review or randomised trial, with its own figures.
  • mixed evidence The cited studies disagree or found limited benefit. Shown because a mixed result is part of the evidence.
  • cited, weaker design Cites an observational study or guidance only.
  • lead Found automatically by the loop and checked only mechanically. A place to look, not a conclusion.

The label comes from the strongest design the entry cites and from whether its sources agree. It is a transparent rule, not a judgement of quality: a trial can still be small, and one setting may not transfer.

4. How sources are checked

Every day the loop searches Europe PMC for the cited paper and confirms that the quoted numbers appear in its abstract. When it finds the paper it records the DOI and PubMed ID, and the entry links to them. Results are shown on the entry as verified, mismatch or not found; a mismatch is shown, never hidden.

5. How implementable is it

Information is only useful once somebody acts on it, so every solution is also scored for implementability from four facts shown on its page. Score 3 to 8 = cost to start (low 3, medium 2, high 1) + real-world adoption (scaled 3, piloted 2, trial-only 1) + an existing delivery channel (2). 7 or more is ready to implement; 5 or 6 is implementable with effort; below 5 is held back as not yet practical. Facts and scoring are in code.

6. How the loop iterates

  1. Refresh the problem ranking from World Bank data (every 6 hours).
  2. Re-check the stalest solution citations.
  3. For each problem without a solution, and for the top-ranked problems, search for recent Cochrane systematic reviews on mortality and survival.
  4. Find bright spots: countries doing better than their income predicts in two separate periods. Leads only.
  5. Quote results sentences from reviews, unchanged, and have a separate step confirm every quote and number against the abstract.
  6. Record every change in the public changelog and feed.

7. What it will not do

It does not present a lead as a finding, paraphrase a result into a stronger claim, or hide a failed check. Corrections: every entry has a Report an error link.

8. Accessibility and speed

The site works without JavaScript, targets WCAG 2.2 AA, and is tested for contrast, structure, links and size on every change. Report anything that does not work for you using the correction link on the page.