This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Slow craft may react too late. Each could judge neighbours by where both will be when its command acts: look ahead by its reaction time.
What was tested
A new switch looks 0–3 reaction times ahead: calm wing, then 5 m/s wind and 1,600 craft. Then 2 reaction times on 1,720 runs of two fixed test sets (mixed, and storm-and-poor-link), 4 craft types.
What was found then
Calm wing losses fell from 532 per 12 runs to 156 at 1 reaction time and 50 at 1.5, then held at 38–42 while close-call time kept falling (69 % to 28 %). In wind, losses fell 102 → 22. At 1,600 craft they fell 73 % but stayed a catastrophe. On the mixed set, wing collisions fell 145 → 5 and eVTOL collisions 23 → 0. The multirotor and the wing's storm set did not measurably move; the ideal craft (instant response) was unchanged to the digit.
What we learned later
Narrowed: the look-ahead also pushes craft apart harder, not only earlier.
research/envelope.md:2927
The test-set statistics were recomputed with the correct method for small samples, and two group-level 'improvements' became too uncertain to call. Craft lost had not been counted on the storm set. The conclusions stood.
research/envelope.md:3056
At 2 reaction times, one storm scenario got worse (see next step).
research/envelope.md:3110
The '4× dense' grid is capped at 1,600 craft, 2.67× the normal fleet.
research/envelope.md:288
What changed
Nothing yet. The look-ahead switch was added, off by default; it was adopted at one reaction time in cycle 85.
Numbers and source references
- 532 → 244 → 156 → 50
- calm wing losses per 12 runs at 0 / 0.5 / 1 / 1.5 reaction timesresearch/envelope.md:2927
- 38 / 42
- losses at 2 / 3 reaction timesresearch/envelope.md:2998
- two
- separately tested parts of the avoidance calculationresearch/envelope.md:2998
- 69 % → 28 %
- fleet breach occupancy from 1.5 to 3 reaction timesresearch/envelope.md:2998
- 102 → 22
- wing losses in 5 m/s wind, 0 vs 1.5 reaction timesresearch/envelope.md:2998
- 1,600 / 6,604 → 1,792
- craft in the dense cell and its losses (−73 %)research/envelope.md:2998
- 145 / 70
- mixed and storm-and-poor-link scenariosresearch/envelope.md:3056
- 4
- craft types in the test collectionsresearch/envelope.md:3056
- 1,720
- runs in the test-collection gateresearch/envelope.md:3056
- 145 → 5
- wing mixed-scenario collisionsresearch/envelope.md:3056
- 23 → 0
- eVTOL mixed-scenario collisionsresearch/envelope.md:3056