A small satellite warms in sunlight and cools in Earth’s shadow every orbit, so whether a reading is normal depends on where in that cycle it was taken. For each channel it learns the typical value at each point of the cycle (minutes since the satellite last came out of shadow), corrects for longer or shorter shadows through the seasons, and compares the last two days with the three weeks before. The curve is a smooth fit around the orbit (a few harmonics, robust to odd frames). Evidence is counted by pass, not by frame, because frames from one pass aren’t independent. Every shift comes with a 90% range from resampling whole passes, and a drift is dated to when it most likely began. Because it learns only from data more than two days old, a new install shows “needs older data” until it has filled in some history, which it does by itself a little at every fetch. Select a row for its orbit profile.

Loading…

Uptime counters give the exact moment: frame time minus uptime. The orbit then says where the satellite was. Radiation upsets cluster over the South Atlantic Anomaly; if restarts land there far more often than the time spent there would explain, that is the signature. Boot counters catch the rest, with the time known only to between two frames.