This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Two broad sweeps put most remaining impacts at storm strength 10 or 14, on a link sending 1 message a second with a 2 s delay.
What was tested
The worst case was rerun with extra recording. Then a wider margin, allowing for gust spread over the age of the news, was tried on the five storm-and-slow-link cases across fourteen families.
What was found then
Every impact had one craft in a gust 10–30 m/s beyond its top airspeed, both craft at full acceleration for 3 s, on 2–3 s-old pictures. Each craft had its own independent gusts, so neighbours 15 m apart often felt 10 m/s different winds. At 4 m/s², nothing can undo that within a 2 s delay. The wider margin lost (127 to 141 on one layout): holding wider in a dense grid creates more conflicts.
What we learned later
Shared gusts (2026-09-07) removed most of these losses (three layouts 442 → 195). Fixes on 2026-09-08/09 to the gust field's edges and to the neighbour search cut about a sixth more (one layout 62 → 52).
README.md:1010, :1377; git bf846a3, 7b00f6c
What changed
Nothing (the wider margin was rejected).
Numbers and source references
- 27 impacts in 504 rows, 19
- cross sweep 3, and those on the storm-link rowREADME.md:830
- 10–30 m/s
- gust above the envelopeREADME.md:830
- 3 s
- time at full acceleration before impactREADME.md:830
- 2–3 s
- view ageREADME.md:830
- 15 m
- neighbour spacingREADME.md:830
- 10 m/s
- typical wind differenceREADME.md:830
- 4 m/s²
- acceleration limitREADME.md:830
- 2 s
- link delayREADME.md:830
- 127 → 141
- seed 0 of the storm-link setREADME.md:830; git e9ce3ca
- 6 → 13
- seed 1, storm 10 rowREADME.md:830
- fourteen
- familiesREADME.md:830
- five
- storm-and-link rowsREADME.md:830