Everything behind the front page, for anyone who wants to check our work.
Outcomes count individual planes: lasting trail predicted and its own trail seen / predicted but not seen / predicted none but its trail lasted / predicted none and none seen. "Untraced lines" are trail-shaped lines no tracked plane explains.
| Hour | Air over the area | Planes predicted to leave lasting trails | Plane outcomes | Untraced lines | |
|---|---|---|---|---|---|
| Fri 11 PM | Borderline | 0 | 0 / 0 / 0 / 11 | 0 | |
| Fri 9 PM | Borderline | 0 | 0 / 0 / 0 / 23 | 0 | |
| Fri 8 PM | Borderline | 0 | 0 / 0 / 0 / 40 | 0 | |
| Fri 7 PM | Borderline | 0 | 0 / 0 / 0 / 17 | 0 | |
| Fri 6 PM | Borderline | 0 | 0 / 0 / 0 / 0 | 0 |
The weather satellite's pixels are about 2 km wide, so it misses thin, fresh trails. Some "predicted but not seen" results are expected for that reason.
Dallas has busy, parallel airways, so a line in the sky can sometimes fall along the path of a plane that did not make it. To measure how often that happens, every satellite image is also matched against the flights from exactly 24 hours earlier (leaving out the same flight numbers and aircraft). Those planes cannot have made today's lines, so every match they get is a tracing error.
Humidity measured against ice, at airliner cruise height. 100% or more means ice crystals survive and trails last.
| Source | Overhead | Anywhere within 240 km | Votes? |
|---|---|---|---|
| European ECMWF | 88% | 100% | yes |
| NOAA GFS (global) | 87% | 100% | yes |
| NOAA HRRR (3 km, hourly) | 67% | 103% | yes |
Measured once on 223 satellite images in which experts marked every contrail by hand (Google's OpenContrails set), images the detector was never tuned on:
To be used, it had to flag images with trails at least 30 points more often than images without, and at least 50% of its lines had to be real trails. It managed 7 points and 12%. Result: failed.
Plainly: this bar was written on 2026-10-23, after this detector's results were known, so it was not set in advance for it. It can only switch the satellite comparison off, never on. It is now locked, and stamped into Bitcoin, for any replacement detector before that detector is tested (the stamped bar).
Being trained once on Google's expert-labelled training images: stopped by an error, retrying: RuntimeError: band names not found; record has ['brightness_temperature_difference', 'cloud_top', 'co2', 'col_min', 'col_size', 'data_10um', 'data_11um', 'data_12um', 'human_individual_masks', 'human_pixel_masks', . See its plan, stamped before training.
A quiet control area in West Texas is checked with the same detector to see how often natural cloud alone makes trail-shaped lines: in the last 24 hours, 5 untraced line(s) in 40 images (5 lines in all).
No chemistry results yet. See the comparison plan, as stamped.
Version 1 was stamped on 2026-10-24 at about 00:02 Central. After an outside review the same night, and before any chemistry had been downloaded (the database held 0 EPA rows when the job was stopped), it was replaced by this version 2, which adds the fake-date check, ground-wind groups, week-block shuffling, longer delays, detection limits, and waiting for Bitcoin to confirm the plan first. Both versions are published (version 1).
What a result can and cannot show. Particles released at cruise height mostly take days to weeks to reach the ground and land spread over hundreds of miles, and about the same number of planes fly over Dallas every day: a lasting-trail day means the upper air let trails last, not that more planes flew. A finding would mean more of a metal at ground level after lasting-trail weather than after dry weather, beyond what dust, rain, wind and season explain, and it must also pass a fake-date check (the same test with labels taken from a week or two later, which cannot have caused anything). It would not by itself show where the metal came from. No finding would mean this monitor shows no such difference; it would not rule out releases whose fall-out lands elsewhere or is too small to measure.
| Source | Successful runs | Last run |
|---|---|---|
| aircraft_db | 1 of 1 | 7:34 PM |
| ecmwf | 26 of 26 | 1:07 AM |
| epa_aqs | 0 of 1 | 12:47 AM |
| gfs | 26 of 26 | 1:07 AM |
| gfs_archive | 1 of 1 | 12:47 AM |
| goes | 44 of 44 | 1:07 AM |
| goes_wv | 40 of 40 | 1:07 AM |
| hrrr | 26 of 26 | 1:07 AM |
| metar | 26 of 26 | 1:07 AM |
| opensky | 10 of 10 | 7:45 PM |
| opensky_control | 34 of 34 | 1:07 AM |
| opensky_main | 77 of 77 | 1:07 AM |
| rain_archive | 1 of 1 | 12:02 AM |
| raob | 26 of 26 | 1:07 AM |
| surface_archive | 2 of 2 | 12:59 AM |
| viirs | 47 of 47 | 1:07 AM |
A fingerprint of the prediction record is stamped into the Bitcoin blockchain as soon as new predictions are locked, and every hour together with the satellite results. The full prediction record, the satellite results and the stamp files are published here, so you can check for yourself that none of them was rewritten later: the records and how to check them.