This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Streams without lanes held no set height, so every vertical dodge was a random step. The three-stream family carried 85 of 210 sweep impacts.
What was tested
First a step rule: 40 m of height for every 120° of heading. Then a smooth rule: 240 m for a full turn of heading. Both were tested on a broad sweep, on short runs at several crossing angles, and on a new scenario with streams crossing at a slant.
What was found then
The step rule cut the sweep from 210 to 113 (three-stream family: 85 to 5). But streams in the same 120° step stayed at one height, and crossings at 30° to 116° still gave up to 35 impacts. The smooth rule took every crossing angle to 0 and the slanted scenario from 86 to 1. The sweep fell further, to 107.
What we learned later
On 2026-09-08 the rule was limited to traffic on fixed headings: a craft turning through east would get a 240 m height jump, never tested. The looping test worlds used to set it up were also affected by the gust-edge bug fixed that day.
README.md:1244; git 9d9b4a1
What changed
Height by heading, 240 m per full turn (the step version was replaced the same day); slanted-crossing scenario added to the full test set (435 cases).
Numbers and source references
- 85 of 210
- three-stream share of cross-sweep impactsREADME.md:1142
- 40 m
- per 120° sectorREADME.md:1142
- 240 m
- per full turn of headingREADME.md:1165
- 210 to 113
- cross sweep 5, sector ruleREADME.md:1142; git dae591a
- 85 → 5
- three-stream familyREADME.md:1142
- 30–116°
- probe crossing anglesREADME.md:1165
- 7 / 24 / 35 / 34
- probe impacts with the sector ruleREADME.md:1165
- 86 to 1
- skew rows over three seedsREADME.md:1165; git c3a7cc1
- 113 → 107
- cross sweep 5, continuous ruleREADME.md:1165; git bc0774e
- 0
- impacts at every probe angleREADME.md:1165