This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Results so far used 600 craft. Does the delay cost hold with fewer and more? Also: what separates colliding pairs from survivors?
What was tested
72 calm runs at 300 and 900 craft, with and without delay. Another 48 re-flown runs recorded which way the heard position was wrong and how hard craft were told to dodge.
What was found then
The link cost held at every density and on every layout: a 1 s delay lost 3.5× (300 craft) and 4.5× (900) more than perfect information. Tripling traffic multiplied losses about six times. Nothing tested was loss-free. Colliding pairs more often heard the other closing slower than it was, and were told to dodge less. But these differences did not grow with delay, so the tests set in advance said they do not explain the delay effect. Fatal pairs increasingly met at shallower angles.
What we learned later
Withdrawn as a rule-out: a difference that does not grow with delay can still be a cause. The raw error gap between fatal and surviving pairs did grow (0.32 → 0.68 m/s). And the dodge measured was what the steering asked for, not what the aircraft actually did.
research/design/s5_post_density.md:74 (Reading, c261 note)
'Six times the losses for three times the traffic' was broken down: with perfect information it comes from almost seven times more meetings, each about as deadly as before.
research/design/s5_ending_share.md:39
What changed
Nothing. A recording tool was added; it is off by default and runs are identical.
Numbers and source references
- 72
- runs at 300 and 900 craftresearch/design/s5_post_density.md:44
- 48
- re-flown runs recording error direction and dodgeresearch/design/s5_post_density.md (Spec Part A) and :74
- 3.5× / 4.5×
- 1 s link vs perfect information at 300 / 900 craftresearch/design/s5_post_density.md:100
- 1.7–2.0×
- 2 s vs 1 s delay at 900 / 300 craftresearch/design/s5_post_density.md:100
- about six times
- loss rate increase for 3× traffic, perfect information (5.97×)research/design/s5_post_density.md:100
- 0.019 → 0.114
- lost per craft-hour with perfect information, 300 → 900 craftresearch/design/s5_post_density.md:100
- 47° → 36°
- median meeting angle of fatal pairs, perfect information → 2 sresearch/design/s5_post_density.md:74 (Reading)
- 7.8–10.9 m/s
- closing-speed range of the pair comparisons (7.77–10.94)research/design/s5_post_density.md:74