This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
The remaining slow-link collisions were lane crossings at 30 m/s, closing at 56–62 m/s on news 2.3–2.7 s old. A dodge chosen on one message is out of date by the next.
What was tested
A craft closing on a heard neighbour faster than it could cross its avoidance range in one delay plus one message gap now slows, to no less than a quarter of its speed. Then we added a release: slow only while the closest point lies beyond the next message.
What was found then
The first version cut one sweep from 252 to 222 collisions over three layouts, and fast impacts (closing at 40 m/s or more) from 61 to 4. But it made the spacing map worse (67 to 80). With the release, the spacing map fell to 51 and the sweep to 210. Storms on a slow link barely moved (186 to 188); reference runs were unchanged.
What we learned later
The closing speeds quoted in this period were overall relative speeds, not the speed along the line between the two craft; cycle 3 found they overstate impact energy roughly 3 times.
research/envelope.md (What cycle 3 found)
What changed
Slowing on slow links, released on committed crossings (adopted 2026-09-08).
Numbers and source references
- 30 m/s
- cruise speedREADME.md:1091
- 56–62 m/s
- closing speedREADME.md:1091
- 2.3–2.7 s
- view ageREADME.md:1091
- a quarter
- minimum speed factorREADME.md:1091
- 252 to 222
- cross sweep 5, three seedsREADME.md:1091; git 2a19228
- 61 to 4
- impacts at 40 m/s closing or moreREADME.md:1091
- 40 m/s
- closing-speed threshold for 'fast'README.md:1091
- 67 → 80
- spacing map with the first versionREADME.md:1091
- 51
- spacing map with the releaseREADME.md:1091; git a136436
- 210
- cross sweep 5 with the releaseREADME.md:1091
- 186 → 188
- storm-link set with the releaseREADME.md:1091
- three seeds
- seedsREADME.md:1091