The shaded background is not a result. It is the Cramér–Rao lower bound on rf position accuracy evaluated on a grid at 120 m altitude — the best any unbiased estimator could do with this geometry and these clocks, computed before any target exists. Green is under about 25 m; red is beyond 250 m. The cliff at the edge of the node footprint is why detection range and localisation range are different numbers. The modality shown is whichever one this scenario actually depends on: drawing an RF bound for a target that emits no RF would describe a capability that will never be used against it.
| emitter | fixes | p50 | p90 | in 95% | trks | det | loc |
|---|---|---|---|---|---|---|---|
| operator | 54 | 6 m | 14 m | 100% | 1 | 3.4 km | 0.7 km |
| quad | 176 | 8 m | 65 m | 99% | 1 | 8.3 km | 3.5 km |
Covariance calibration: truth fell inside the stated 95% ellipse 97.0% of the time. An honest system lands near 95%; well below means the error bars are too small and every downstream decision inherits that over-confidence.
At the speed of light 1 ns of clock disagreement is 30 cm and 1 µs is 300 m. Nothing else in the system matters as much.
| frames sent | 2283 |
| delivered to hub | 2283 |
| relayed (multi-hop) | 1984 |
| dropped by link loss | 97 |
| forged / tampered rejected | 0 |
| replays rejected | 0 |
| nodes able to reach hub | 9 / 9 |
| hub key fingerprint | 7e37bdf90daf1c8b |
Every detection above crossed this mesh — X25519 key agreement, ChaCha20-Poly1305 per frame, authenticated routing headers, sliding-window replay rejection — before fusion was allowed to see it. A frame the mesh loses is a detection the picture does not contain.