CrashAtlas

Aviation accident data exists in public sources, but it is scattered, inconsistently formatted and almost never geocoded. Turning it into something you can explore means solving ingestion and normalisation before you can even think about the interface.

The map is the product, and a map with eighteen thousand markers is a performance problem, an accessibility problem and an SEO problem at the same time: search engines cannot read a canvas.

A normalisation pipeline turns heterogeneous public sources into one record shape — date, operator, airframe, location, cause, fatalities — with geocoding and de-duplication as explicit, reviewable steps.

The map clusters server-side by region and streams only what the current viewport needs, so panning stays smooth at full archive scale.

Every dimension of the data is also a set of ordinary, crawlable pages: by country, by year, by airline, by aircraft, by cause. The map is the experience; the pages are the index — and they are the fallback when the map cannot load.

Two audiences share the site: someone following a specific historical accident, and someone exploring patterns. The map serves the second, the archive pages serve the first, and every marker has a permanent page behind it.

The map's loading state is not a spinner — it explains what the map does and links straight into the browsable archive, so a slow connection never becomes a dead end.