Calibration & reliability
The engine is graded against D-ATIS ground truth at towered fields, with the ATIS deliberately withheld from the prediction. When the engine says 80%, is it right 80% of the time?
| bin | predicted | observed | n |
|---|---|---|---|
| 0.1–0.2 | 0.17 | 0.67 | 401 |
| 0.2–0.3 | 0.25 | 0.66 | 1880 |
| 0.3–0.4 | 0.35 | 0.61 | 3389 |
| 0.4–0.5 | 0.44 | 0.55 | 2386 |
| 0.5–0.6 | 0.54 | 0.69 | 48033 |
| 0.6–0.7 | 0.65 | 0.68 | 40513 |
| 0.7–0.8 | 0.75 | 0.68 | 27776 |
| 0.8–0.9 | 0.85 | 0.66 | 32251 |
| 0.9–1.0 | 0.97 | 0.67 | 50914 |
Brier score: 0.255 · calibration error (ECE): 0.160 · samples: 208209
Per-source health (24 h)
| source | ok | errors | p50 latency | last note |
|---|---|---|---|---|
| ADSB | 28508 | 13975 | 203 ms | |
| ADSB_DEDUP | 501 | 0 | 38 ms | dedup rate 7.7% |
| D_ATIS | 288 | 0 | 213 ms | |
| METAR | 1115 | 21 | 193 ms | |
| NASR | 4 | 0 | — | 01_Oct_2026 current |
| NOTAM | 28779 | 0 | 53 ms | |
| PROC_SOURCE | 21 | 0 | 227 ms |
Airport data cycle
Serving the FAA NASR 01 Oct 2026 cycle; that is the cycle in effect. Checked every six hours. Ingested 2026-10-01 04:12 UTC.
Corpus so far
- 467011 stored predictions (467011 with ATIS withheld)
- 1193790 detected operations across 669 airports
- 129880 archived ATIS broadcasts
- 1 pilot feedback reports