This is an earlier experiment. Read the amendments below before treating its findings as current.

The question

Reviewers asked for a real change. Suspect: the rule that makes each craft match its neighbours' speed and direction.

What was tested

The rule at full, half and zero on the wing (six layouts); then off for other airframes and the wing in calm, walled lanes, double traffic, wind, crowds and a poor link.

What was found then

It is what slows the wing: time at minimum speed fell from 25 % to 3.6 % with the rule off, and stall losses from 83 % to 52 % of the fleet. In calm conditions nothing got clearly worse without it. But in crowds it does real work: slow-responding craft at four times traffic went from 1.65 to 4.63 collisions per 1000 km without it, and crowded wings fared worse. Deliveries rose 13–32 % in crowded and poor-link skies. Switching the rule off by default was ruled out.

What we learned later

LATER / CYCLE 30

The calm-grid 12 % figure is the share of run time with any pair in a close call, not a share per pair.

research/reviews/brief-c31.md

LATER / CYCLE 31

'A speed cost that buys spacing' was taken back: the effect differs by kind of airspace (a dense airport area got safer without the rule, 72 to 26 collisions).

research/reviews/brief-c32.md

What changed

Nothing. A strength setting for the speed-matching rule was added; default unchanged.

Corrections retained in the record

  • Earlier wording: at half strength collisions were higher than at either extreme, in every seed

    Record correction: 70 collisions at half vs 49 and 50; every run above the full-strength median, not above both extremes in every seed

    research/data/oos/wingalign.md:8-27

Numbers and source references
6
seeds per caseresearch/reviews/brief-c29.md
15
airspace types per crowded/link conditionresearch/reviews/brief-c31.md
9 m/s
wind in the wing testresearch/reviews/brief-c30.md
25 % → 11 % → 3.6 %
wing time at its floor, rule at full / half / offresearch/reviews/brief-c29.md
83 % → 52 %
fleet lost to stalls, rule on vs offresearch/reviews/brief-c29.md
1.65 → 4.63
slow craft, 4× traffic, collisions per 1000 km, rule on vs offresearch/reviews/brief-c31.md
0.48 → 0.85
slow craft, 2× traffic, sameresearch/reviews/brief-c31.md
17.8 → 21.4
wing at double traffic, sameresearch/reviews/brief-c30.md
13–32 %
rise in deliveries without the rule (+24 / +23 % crowded, +32 / +13 % link)research/reviews/brief-c31.md
Original structured record ↗