This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
On the wing's storm-and-poor-link test set the look-ahead's effect was unclear. Where does it help, and where does it hurt?
What was tested
280 matched storm-set comparisons at 2 reaction times; re-flights of the 2 worsening scenarios at 0–2, turbulence on and off; then 1 reaction time on both test sets for three slow types (630 runs).
What was found then
Storm-set collisions fell 897 → 796, too little to tell from luck. Six scenario groups improved and one got worse: a 405-craft hovering formation knocked by a storm (24 → 75). On new layouts, 2 reaction times doubled its collisions (62 → 130), 1 showed no clear change, and without turbulence the look-ahead helped (45 → 22). Read then as projected turbulence inventing near-misses. At 1 reaction time, wing losses on the mixed set fell 290 → 64 and eVTOL collisions 23 → 3, read then as keeping its whole gain. No type was measurably worse.
What we learned later
Narrowed: the harm depends on turbulence, but 'turbulence noise fed into the projection' is plausible, not proven. 'Harmless at one reaction time' was withdrawn because the uncertainty still allowed a small rise.
research/envelope.md:3166
Narrowed: the eVTOL kept 20 of its 23 removed collisions (87 %), not its whole gain.
research/envelope.md:3212
What changed
Nothing in flight. Two counting fixes: the storm test now counts craft lost, and a formation's parked reserve craft are no longer counted as lost.
Numbers and source references
- 280
- matched storm-set comparisons (four sets of random starts)research/envelope.md:3110
- 897 → 796
- wing storm-set collisions, look-ahead off vs 2 reaction timesresearch/envelope.md:3110
- six
- scenario groups that improvedresearch/envelope.md:3110
- 405 / 14 m/s
- craft in the kicked hovering formation, and storm strengthresearch/envelope.md:3110
- 24 → 75
- collisions in the kicked-formation groupresearch/envelope.md:3110
- 2 × 10
- worsening scenarios × random starts in the follow-upresearch/envelope.md:3166
- 62 → 130
- their collisions at 2 reaction timesresearch/envelope.md:3166
- 62 → 53
- the same at 1 reaction timeresearch/envelope.md:3166
- 45 → 22
- the same at 2 reaction times with storm turbulence offresearch/envelope.md:3166
- 630
- runs of the one-reaction-time test collectionsresearch/envelope.md:3212
- 145 → 32 / 290 → 64
- wing mixed-scenario collisions / craft lost at 1 reaction timeresearch/envelope.md:3212
- 23 → 3
- eVTOL mixed-scenario collisions at 1 reaction timeresearch/envelope.md:3212