Methodology

Every point on the map is a reportfiled by a person — never a confirmed object. We say "reports cluster here," never "UFOs appear here." That distinction is the whole point.

The strongest predictor of where reports appear is simply where people live and where the reporting culture is strong. Report volume spikes after films, news waves and viral moments. Our main source (NUFORC) is heavily US-weighted, so the map partly draws the United States, not the sky.

Wherever a count is shown, read it as "reports made," not "events that happened."

Because population drives raw counts, the atlas will always offer a normalized view alongside the raw one — reports per 100,000 residents per cell, using a gridded population dataset. Raw dots impress; normalized rates inform. (Normalization view is on the build roadmap; see the plan.)

"Reports within 7 days and 100 km of an M4+ quake" sounds striking until you know how often that happens by chance. The Correlation explorer(in the map controls) compares the observed count against a date-permutation null — every report's location is kept fixed while its date is drawn from the real pool of report dates, so seasonal and decade-to-decade reporting waves are held constant. You get observed vs. expected, a z-score and p-value, and an honest read — for the earthquake/tectonic-strain hypothesis it currently comes out not distinguishable from chance.

A low p-value only means the space-time pattern is unlikely under random dates — never proof of a cause. And some effects need a different lens: wide-area events like Starlink trains are seen across a whole continent, so a "distance from the launch pad" test won't capture them.

NUFORC locations are city-level, not exact coordinates. A dot marks the reporter's town, not a pinpoint. Our own future reports (see the report form) can be more precise and are tagged accordingly.

Most sources don't say whether a report was ever resolved — so a dot tells you nothing about whether it's mysterious. France's official unit, GEIPAN (part of the CNES space agency), does: every case is classified A (identified), B (probably identified), C (insufficient information) or D (unexplained afterinvestigation). The "Official cases" layer shows these, with the D-cases highlighted. They're region-level (≈10 km) and only ~2–3% end up as D — a useful reality check against the raw dot map.

The tool presents; you interpret. It will never tell you what a report "was." It gives you the layers, the counts, the null model and the caveats — and gets out of the way.