This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
On a link past its breaking point, a neighbour last heard seconds ago may have moved. Could craft widen their margin by the age of the news?
What was tested
We added a margin that grows with message age and tested it on 70 the slow-link cases (a link just too slow and lossy to keep up): once for every pair, and once only for pairs moving faster than 2 m/s relative to each other.
What was found then
Used for every pair, it wrecked formations: the shoved hovering grid went from 0 to 84 collisions, because hovering neighbours are already predicted exactly. Used only for pairs faster than 2 m/s, it cut the slow-link cases from 87 collisions to 37. Throughput moved by less than 0.5 %. The full test set went from 8 to 7, with still-air results untouched.
What changed
A margin for old news, used only between pairs moving faster than 2 m/s relative to each other.
Numbers and source references
- 70
- cliff casesREADME.md:366
- 0 → 84
- kick grid collisions with the ungated marginREADME.md:366
- 2 m/s
- relative-speed gateREADME.md:366
- 87 to 37
- cliff set collisionsREADME.md:366; git 2138053
- 56 → 23
- turbulence 3 at 1 HzREADME.md:366
- ±0.5 %
- throughput changeREADME.md:366
- 8 → 7
- full-matrix collisions (run 15)README.md:366