The water we hear
The ships side of the network has two receivers feeding it now. One watches the Lower Mississippi from the Port of New Orleans, the other went up today on the Humber in England. The coverage map showed one of them.
It wasn't a data problem. Both stations had been writing squares all evening, and both were in the payload the page fetches. The map simply framed itself somewhere the second one wasn't.
Framing on a share is framing on the loudest
The coverage map picks its view by finding the squares that carry 99 % of the messages and fitting the frame around those. That rule exists for a good reason: AIS arrives over a noisy VHF channel, a payload that passes its checksum can still decode to a position in the wrong ocean, and one such square would otherwise zoom the map out to the planet.
It works for one receiver. On a network it quietly becomes "frame on whoever talks most". New Orleans delivers around 478,000 messages a day. The Humber station, a few hours old, had delivered 3,300. Ninety-nine percent of the combined total is reached before you leave Louisiana, so six squares of the Mississippi set the bounds and all twenty British ones sat off the edge of the canvas.
The map now frames on any square carrying at least twenty messages, whoever heard it. A share of the total favours the biggest contributor; a floor doesn't favour anyone. Twenty is where the line goes because a transponder repeats its position every few seconds under way, so a real contact leaves dozens of messages in the square it crossed, and a corrupt decode leaves exactly one.
Forty squares of nonsense
Checking the numbers turned up the second fault. Of 71 lit squares, 40 carried a single message each, and they were in Antarctica, at the North Pole, off Kamchatka and in the middle of the Pacific. From two antennas, one on a river in Louisiana and one on an English estuary.
Several of them share a coordinate with a real square: a whole column at longitude −90.05, a row at latitude 29.95. That's the signature of one half of a position decoding wrong while the other half survives. The server now drops any square more than 300 km from the busiest square that same receiver heard, which is generous for marine VHF and merciless on the North Pole. It costs us 45 messages out of 1.6 million.
While we were in there: the page said "from 3 receivers" over a grid two of them drew. The third is our own Hamburg station, registered and never yet fed. It counts the ones that actually put a square on the map now.
No single zoom holds two continents
Even framed correctly, one map of New Orleans and the Humber is the North Atlantic at a zoom where a 0.1° square is smaller than a pixel. Both receivers on screen, neither of them readable.
So the coverage page now draws one map per stretch of water, busiest first. The clusters are geographic and never per receiver (two antennas watching the same estuary land on the same map), because the page promises not to tell you which antenna heard which square, and it keeps that promise.
And the map that was missing
The aircraft side has had two maps for a long time: tar1090 for what is flying right now, and a coverage grid for where the network has heard anything over the last thirty days. The water only ever had the second one. A volunteer whose receiver is working at this moment had nowhere to watch it work.
There is now a live one. The map filtered to our own receivers shows the vessels a Skydex AIS station is hearing, with everything the upstream feed contributes filtered out. As of tonight that's 90 hulls of the 1,314 on the public map: 62 around New Orleans, 27 on the Humber, and all 27 of the Humber ones carry the flag. Until this afternoon that estuary had never appeared in our daily coverage roll at all, so almost everything on it now reaches us because somebody in England put an antenna up.
Which antenna heard which hull is still not published, and won't be. The flag on a vessel says the network heard it. That is the interesting half, and it is the only way a feeder can recognise their own water on a map that otherwise looks the same everywhere.