How it works

1. Measure the air where planes fly

Temperature, humidity and wind between about 23,000 and 45,000 feet come from three independent forecast systems (two from NOAA, one from Europe), checked at 25 points across and up to 240 km upwind of Plano, Texas. The Fort Worth weather balloon and the satellite's water-vapour reading are shown as cross-checks but never decide anything.

2. Work out what each plane's trail should do

For every plane above 23,000 feet, using its altitude and engine type:

• Will a trail form? The Schmidt–Appleman test (Schumann, 1996): is the air cold enough for the exhaust's water to freeze?
• Will it last? Only if the air holds 100% or more of the moisture ice can tolerate. Below that, the ice crystals evaporate within a minute or two.
• Will it spread into haze? If it lasts and the wind changes sharply with height.

A plane gets a firm prediction only when all three forecasts agree.

3. Lock the prediction before looking

Each 10-minute prediction uses only data downloaded before that 10 minutes began. It is written to a record where every entry is chained to the one before, so changing any past entry breaks the chain, and the record is stamped into the Bitcoin blockchain right away. Only then is the satellite checked.

4. Look at the sky

Every 5 minutes the GOES-East weather satellite photographs Texas in infrared. A fixed detector finds trail-shaped lines. Each line is traced back to the plane that made it, allowing for the wind that moved it and the satellite's slanted view (about 11 km over Plano). The detector's settings were calibrated once on expert-marked images from elsewhere, then locked; they are never tuned on this sky.

5. Compare, honestly

• Every plane's prediction is scored against its own trail, never against lines in general.
• The skill score is built so that always guessing "trails" scores zero.
• A trail that lasted in dry air counts as unusual only beyond what coincidence would produce. Coincidence is measured every day by matching the same lines against the previous day's flights, which cannot have made them. Those stand-in flights compete with fewer planes for each line than the real ones do, so this slightly overstates coincidence, which makes the test lean cautious rather than eager.
• All results are shown, including those that agree with the standard explanation.

6. What is in the air on trail days

A satellite or camera can show whether ice is there and how long it lasts, not what a trail contains. So the chemistry comes from EPA's own monitor in Dallas, which tests the air for metals every third day. Each of those days is labelled from the weather alone: lasting-trail air over the city for at least 12 of the 48 hours ending that day, or none at all. Aluminum, barium and strontium are compared between the two kinds of day, each divided by silicon so that ordinary wind-blown dust (which contains all three) does not decide the answer, and only within the same month, the same rainy or dry weather and the same ground-wind direction. Whole weeks are shuffled to judge chance, because weather patterns last several days. The same test is also run with labels from a week or two later, as a fake-date check: if that also "finds" something, it is weather, not trails. A metal that EPA's instrument mostly cannot detect is reported as not testable. The whole plan was written down and confirmed in the Bitcoin blockchain before any chemistry was read.

What this can and cannot see

• The weather satellite has 2 km pixels: it sees trails once they widen, not fresh thin ones.
• It recognises trails by the infrared signature of ice. A planned roof camera will record every visible trail, whatever it is made of.
• Planes without transponders are not tracked, so a line nobody made may come from such a plane.