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

The question

The last test hinted that the pull back toward the lane (a gentle steering command) itself costs collisions in crowded skies.

What was tested

Six layouts each on the busy sky at four times traffic, slow-responding craft: pull at full, 43 %, 14 % and off. We also tried removing only its vertical part.

What was found then

Collisions fell as the pull weakened, from 3.9 to 1.6 per 1000 km, with deliveries unchanged; the two ends are clearly apart. But the sky stayed unsafe at every strength: some pair was in a close call nearly all the time, and the closest approaches were no milder. On the calm crossing grid, ideal craft (they respond instantly) went from 0 to 3 collisions without the pull. Removing only the vertical part did not help. A weaker pull is a cheaper way of not having lanes, not a fix.

What we learned later

LATER / CYCLE 23

The cost was traced to the sideways part of the pull back toward the lane.

research/reviews/brief-c24.md

What changed

Nothing.

Corrections retained in the record

  • Earlier wording: the “switch the pull-back off” test cut … the busy grid’s by half

    Record correction: the halving (3.93 → 1.97 per 1000 km) was on the calm 1× crossing grid, not a busy grid

    research/reviews/brief-c20.md

  • Earlier wording: Every step is worse than the last on that count

    Record correction: collisions fall at every step (3.93 → 2.77 → 2.13 → 1.59); adjacent intervals overlap, ends apart

    research/reviews/brief-c21.md

Numbers and source references
6
seeds per variantresearch/reviews/brief-c21.md
4×
traffic multipleresearch/reviews/brief-c21.md
3.9 → 2.8 → 2.1 → 1.6
collisions per 1000 km with pull-back cap 0.35 / 0.15 / 0.05 / ~0 of cruiseresearch/reviews/brief-c21.md
43 % / 14 %
the 0.15 and 0.05 caps as shares of the default 0.35research/reviews/brief-c21.md
99.6–99.9 %
run time with some pair inside the standard, at every settingresearch/reviews/brief-c21.md
0 → 3
ideal craft collisions on the calm crossing grid, pull-back offresearch/reviews/brief-c20.md
4.6
collisions per 1000 km with only the vertical pull-back removed (4.63)research/reviews/brief-c22.md
−41 %
fewer close crossing moments from full to no pull-backresearch/reviews/brief-c21.md
Original structured record ↗