This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Traffic between the normal and the crowded grid, and its mix with wind, was unflown for slow craft.
What was tested
Grids of 600–1,600 craft for three slow types, with and without the look-ahead, in calm and (wing only) 5 m/s wind; the wing at 1,200 craft in 0–5 m/s wind: 1,172 new runs.
What was found then
With the look-ahead, the wing's collision rate roughly doubles per 300 extra craft (22 to 173 per 1,000 craft-hours from 600 to 1,600). No level was safe; the look-ahead was worth one traffic step. 5 m/s wind lowers the rate at every level. Close calls mostly involve one pair; 97 % end with the pair separating, slightly sooner with the look-ahead. Calm lanes at 1,200 craft deliver 54–64 % of their no-traffic laps; 3 m/s wind lifts that to 79–97 %. In 30 min, 8.8 % of the fleet was lost in calm, 2.2 % in 5 m/s wind.
What we learned later
'Exactly one step, not two' was dropped because it was read off neighbouring numbers. Tested in cycle 91: one step for wing and multirotor, unclear for eVTOL.
research/envelope.md:3440; research/envelope.md:5657
Narrowed: 'one pair at a time' fits the counts but cannot rule out a few long chains; earlier ~1 s durations had counted runs with no close call as zero.
research/envelope.md:3487; research/envelope.md:5709
The idea of 'a standing jam at crossings that wind breaks up' was dropped: no pile-up was found at crossings or lane ends.
research/envelope.md:5614
All these wind runs used one wind direction; in stage 5, wind from 70° did not reduce losses consistently.
research/envelope.md:8592-8600
What changed
Nothing. Counting-only recording switches were added, off by default.
Numbers and source references
- 600 / 900 / 1,200 / 1,600
- fleet sizes on the calm gridresearch/envelope.md:3440
- 1,172
- new run records across cycles 88–90 (144 + 502 + 526)research/envelope.md:3440; research/envelope.md:3487; research/envelope.md:5709
- 22 / 47 / 94 / 173
- wing collisions per 1,000 craft-hours with the default, calmresearch/envelope.md:3440
- 4.5 / 12 / 23 / 36
- the same in 5 m/s windresearch/envelope.md:3487
- one step
- density steps the look-ahead is worth (cycle 91)research/envelope.md:5657
- 97 %
- pair breach episodes that end with the pair separatingresearch/envelope.md:5709
- 2.56 → 2.25 s
- mean of run-median durations, separation-ended, mixed scenarios, old vs newresearch/envelope.md:5709
- 54–64 %
- share of free-flow laps each lane delivers in calm at 1,200 craftresearch/envelope.md:5657
- 79–97 %
- share of free flow at wind 3 m/sresearch/envelope.md:5657
- 8.8 % / 2.2 %
- share of a 1,200-craft wing fleet lost in 30 min, calm / wind 5research/envelope.md:5709