In October 2012, geologists aboard the Australian research vessel RV Southern Surveyor set out on a 25-day voyage across the eastern Coral Sea to study the region's plate tectonics. On a transit leg between two dredge sites, the ship passed within a few miles of a feature carried by Google Earth, by the global seafloor and coastline databases used across the earth sciences, and by the mapping software built on top of them: an island 25 kilometers long and 5 kilometers wide, labeled Sandy Island. The ship's own navigation charts, the actual legal charts for the area, showed none of that. They showed open water, and had for decades.
The island had entered the record in 1908, when a British Admiralty chart recorded an island the whaling ship Velocity's master had reported sighting in 1876, roughly 500 kilometers northwest of New Caledonia. Hydrographers never confirmed it, and eventually unmade it. The French Hydrographic Service struck Sandy Island from its own charts in 1974; the Australian Hydrographic Service followed in 1985, after its own seafloor survey found nothing shallower than 1,488 meters where the island was supposed to lie. Both services passed the correction to other national hydrographic offices. As far as the people who actually draw charts were concerned, the matter was closed.
The correction never reached the data everyone else was using. Around 1990, an American mapping agency digitized a global set of paper charts into the World Vector Shoreline Database, a single coastline any researcher's software could read, and the uncorrected island went in, labeled as land. The scientists who later wrote up the mystery attributed errors like it to a mix of mistakes in the digitizing and mistakes already sitting in the charts being digitized. Every later product built on that database, a shoreline standard still used across the earth sciences, Google Earth included, inherited the island without anyone checking its coordinates against an actual chart or a sounding.
Steven Micklethwaite, a geologist aboard the Southern Surveyor, joked afterward that the crew had a good laugh at Google as they passed the spot where the island was supposed to be. Within a week, researchers working independently on mailing lists and online forums had compared the available satellite imagery and bathymetric surveys and reached the same conclusion the hydrographic services had reached decades earlier: there was no island, and the data had never supported one.
A correction made once, at the source, is not the same as a correction that reaches every place the error was copied. Hydrographers fixed their own record in 1974 and 1985; Sandy Island lived on anyway, in the scientific coastline and bathymetry databases and the satellite imagery built on top of them, because something corrected once still had to be corrected everywhere it had already been copied. The fix is not finished when somebody writes it down. It is finished when the copies catch up.