The world got less murderous, and one country shows what it can cost

Status: drafted 2026-08-13 Reachability swept first: homicide-rate-unodc and prison-population-rate both fetch clean; road-deaths-per-100000, press-freedom-index and democracy-index-eiu (403) do not. Vintage checked: UNODC via OWID through 2024; World Prison Brief through 2026. Both current. Note the two end in different years — see Q4.

Segment: Global homicide fell a quarter this century, and almost nobody believes it

• Source theme(s): Batch 39 (crime). The bank's own notes record crime as its single largest hole — zero real coverage across 884 prior themes, so this is the first artifact on the shelf in the domain.
• Spine role: main story
• The framing people carry: Violent crime is rising, cities are more dangerous than they were, and the world is getting more violent.
• Hypothesis under test: That the perceived rise in violence is visible in the homicide record.
• Verdict: REFUTED on the aggregate, with a much more uncomfortable finding underneath. The world homicide rate fell from 6.9 to 5.1 per 100,000 between 2000 and 2024 — down 25%, and falling in 20 of 24 years. What survives is a composition story (the decline is wildly uneven) and a price story that the aggregate hides completely.

The numbers (every row sourced — no exceptions)

Homicide: UNODC via OWID grapher homicide-rate-unodc, accessed 2026-08-13. Incarceration: World Prison Brief via OWID grapher prison-population-rate, same date. Both Tier A (existing OWID connector). All figures per 100,000 people.

The texture (what narrative flattens)

• The lead finding: the aggregate fell and the perception didn't. 6.9 → 5.1 per 100,000, declining in 20 of 24 years. This is one of the largest improvements in human safety of the century and it is essentially absent from public understanding of crime.
• The decline is almost entirely NOT in the region people picture. Europe fell 69% — from 7.8, above the world average in 2000, to 2.4. The Americas fell 11% and still sit at 14.0, nearly three times the world figure. The global improvement is mostly Europe and Asia; the region whose violence dominates coverage barely moved.
• The price, and it is the part that stops this being a good-news story. El Salvador went from the most murderous country in the record — 138.4 in 1995, still 107.6 in 2015 — to 7.9 in 2022, a 93% collapse in seven years. Over the same period its prison population rate went from 617 to 1,659 per 100,000: about 1.6% of the entire population incarcerated, three times the United States at its own historic peak. The two series are the same argument read from opposite ends.
• Meanwhile the United States went the other way on both counts — incarceration down 28% from its 2008 peak and homicide down 38% since 1990. Whatever the relationship is, it is not a simple one, and the US is the counter-example to any clean reading of El Salvador.
• Brazil is the null case: homicide essentially flat over three decades (19.6 → 19.3) while its prison rate more than tripled (132 → 416).

Datoid artifact

• Dashboard homicide, 4 panels: the world line; the regional decomposition; El Salvador's two series on one axis; and the US/Brazil/Colombia contrast. Both metrics are rate per 100k, so they share an axis by design — that shared scale is what makes the El Salvador panel legible.
• First crime artifact on the shelf. development category (crime is not security, which is defence/arms-control).

Story justification (Stage 1 rule 5)

• Reveal that needs sequencing? YES → build.
• Because: the punchline is a second dataset, and it must be off-screen while the trap is set. The sequence is: world homicide fell a quarter → the fall is concentrated in Europe, not the Americas → except one country in the Americas fell further than anywhere on earth → so what did that take? → reveal the incarceration line. Showing both from frame one turns the strongest editorial moment in the batch into a footnote.
• Slug: /stories/homicide. Deferred if time-boxed; recorded as owed.

Open questions / where I'm guessing

These must NOT become spoken assertions.

1. No causal claim about El Salvador is supported by this data, in either direction. The mass-incarceration policy is contemporaneous with the homicide collapse; contemporaneity is not causation, gang truces and demographic change are also in the period, and the human-rights dimension (detention without trial under a prolonged state of exception) is a documented controversy this dataset says nothing about. Present the two series and the timing. Do not say the policy worked, and do not say it didn't.
2. Homicide statistics are recorded crime, and recording quality varies enormously by country and improves over time. A falling rate can partly be a changing count. UNODC harmonises, but not perfectly.
3. African coverage is the weakest in the set — many countries estimate rather than count, so the regional line carries the widest error bars. Use it for the shape, not for a precise figure.
4. The two series end in different years (homicide 2024 world / 2022 El Salvador; prison 2024–2026). Never compute a ratio across mismatched years or imply a point-in-time pairing that the data does not contain.
5. Prison population rate is a stock, homicide a flow. They are not commensurable quantities despite sharing a per-100,000 denominator; the chart puts them on one axis because the scale suits it, not because the ratio means anything.
6. "Nobody believes it" is an assertion about public perception that I have not sourced. Either cite a real survey or phrase it as the framing the segment is testing, not as a finding.