This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Removing only the vertical part of the pull back toward the lane did not help. But both parts shared one speed limit, so the sideways part could not be judged alone.
What was tested
We measured sideways offsets and how often the pull hits its limit, then gave each part its own limit and switched the sideways part off (six layouts, predictions written in advance).
What was found then
Lanes do not hold crowded craft sideways either: eastbound craft sit a typical 171 m from the centre of a 240 m-wide lane, with the pull mostly at its limit. With the sideways pull off, collisions drop to 1.5 per 1000 km, as with no pull at all; with both parts on under separate limits, 4.4. Across five versions, collisions follow the sideways push. That is a pattern, not a proven cause, and the sky stays unsafe throughout.
What we learned later
The line of work was closed; the planned test of weaker and stronger sideways pulls was never run, so how it costs collisions stays open.
research/reviews/consensus-c24.json
What changed
Nothing. Separate limits for the two parts of the pull were added as a test switch, off by default.
Numbers and source references
- 6
- seedsresearch/reviews/brief-c24.md
- 240 m
- lane widthresearch/reviews/brief-c22.md
- 171 m / 131 m
- median sideways offset, eastbound / express laneresearch/reviews/brief-c23.md
- 64 % / 54 %
- share of samples outside the 120 m half-width, eastbound / expressresearch/reviews/brief-c23.md
- 87 % / 76 %
- share of samples with the pull-back at its cap, two lanesresearch/reviews/brief-c23.md
- 1.5
- collisions per 1000 km, sideways pull off (1.47)research/reviews/brief-c24.md
- 4.4
- collisions per 1000 km, separate-budget control (4.41)research/reviews/brief-c24.md
- 3.9–4.6 vs 1.5–1.6
- collisions per 1000 km with vs without the sideways push, five variantsresearch/reviews/brief-c24.md