This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Gusts had been independent for each craft, the gloomiest extreme. Real gusts are shared over hundreds of metres.
What was tested
We built a gust field shared by craft in the same patch of air, with a patch-size setting (0 means the old independent gusts). It ran on the storms on a slow link cases with patches from 0 to 120 m, then on three layouts at 60 m.
What was found then
Most of those losses came from the independent-gust assumption. On one layout, collisions fell from 150 to 63 with 60 m patches; larger patches helped slightly more and smaller ones less. Over three layouts at 60 m they fell from 442 to 195, with work up 0.5 %. The owner chose 60 m as the realistic default. The full test set stayed at 0, with 342 of 406 cases exactly unchanged.
What we learned later
The 2026-09-08 outside review found the gust field did not wrap around at the edges of the looping test worlds, and the neighbour search was limited by airspeed. Both were fixed, and these losses fell about a sixth more (one layout 62 → 52).
README.md:1244, :1377; git a17e672, 7b00f6c
What changed
A shared gust field, with 60 m patches by default since 2026-09-07 (0 restores independent gusts).
Numbers and source references
- 0, 30, 60 and 120 m
- cell sizes triedREADME.md:1010
- three seeds
- at 60 mREADME.md:1010
- 150 → 78 / 63 / 60
- seed 0 at 30 / 60 / 120 m cellsREADME.md:1010; git bf846a3
- 442 → 195
- three seeds at 60 mREADME.md:1010
- +0.5 %
- workREADME.md:1010
- 60 m
- default cell sizeREADME.md:1010; git 2a02d95
- 342 of 406
- matrix rows bit-identical at the defaultREADME.md:1010