This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Before adopting it: does one reaction time hold on the wing's own map conditions and on fresh layouts?
What was tested
The wing's map conditions at one reaction time; both test sets on new layouts for wing and eVTOL (860 runs); then adoption and a first timing of close-call episodes.
What was found then
Wing losses fell 532 → 156 per 12 runs in calm and 102 → 32 in 5 m/s wind. At 1,600 craft they fell 6,604 → 2,992, less than 1.5 reaction times had achieved. Fresh layouts repeated the gains: pooled, wing losses on the mixed set fell 554 → 132, and storm-set collisions 454 → 382, now clear of luck. The knocked hovering formation did not change. Adopted 2026-09-14; the ideal craft (instant response) is unchanged to the digit. Wing close calls became fewer (6.5 → 4.7 per craft-hour), shorter and less often very close.
What we learned later
Narrowed: the change in the share of long close calls was measurable but uncertain, not 'impossible to measure'. 'Fewer very close episodes' means a lower rate, not shallower ones.
research/envelope.md:3324
The typical length of about 1 s had counted runs with no close call as zero. In runs that had any, close calls ending with the pair moving apart typically last about 2.5 s, and the look-ahead shortens them by about 0.3 s.
research/envelope.md:5709
What changed
The look-ahead became one reaction time by default for every craft with a response lag (2026-09-14). Craft with no lag are unaffected.
Numbers and source references
- 24
- runs on the wing's map conditionsresearch/envelope.md:3254
- 102 → 32
- wing losses per 12 runs in 5 m/s windresearch/envelope.md:3254
- 6,604 → 2,992 (1,792)
- wing losses at 1,600 craft at 1 reaction time (1.5)research/envelope.md:3254
- 532 → 156
- calm wing lossesresearch/envelope.md:3254
- 860
- runs on the second set of random startsresearch/envelope.md:3287
- 554 → 132
- wing mixed-scenario craft lost, both sets pooledresearch/envelope.md:3287
- 454 → 382
- wing storm-set collisions, both sets pooledresearch/envelope.md:3287
- 15 → 15
- kicked-formation collisions, pooledresearch/envelope.md:3287
- 2026-09-14
- adoption dateresearch/envelope.md:3324
- 433
- runs in the adoption and episode stepresearch/envelope.md:3324
- 6.5 → 4.7
- wing breach episodes per craft-hour, mixed scenariosresearch/envelope.md:3324
- 1.3 → 0.9 s
- typical (median) episode length at the time (1.28 → 0.89)research/envelope.md:3324
- 0.36 → 0.14
- deep breach episodes per craft-hourresearch/envelope.md:3324