This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Delay matters. What is the shape of the curve, checked on new layouts?
What was tested
84 half-hour runs in calm air on six new layouts, each starting with 600 craft, with perfect information and delays of 0.25–3 s.
What was found then
Losses rose steadily with delay on every layout, from 0.05 per craft-hour with perfect information to 0.71 at 3 s. The 0.5 s and 2 s results matched the earlier layouts. A straight line fits at about +0.22 per extra second. No delay tested was loss-free. Longer delay barely lengthens time in the bubble, but craft get much deeper into it (2.8×). The share of bubble visits ending in a loss rises from 0.25 % to 3.7 %.
What we learned later
The summary was narrowed: all lanes fly in one loosely held layer of sky, not two layers. Only the 0.5 s and 2 s points were repeated on a second set of layouts, so the +0.22 slope is a fit on one set, not a law.
research/design/s5_lat_fresh.md:53 (Map sentence, c256 repair)
Runs start with 600 craft and lost craft leave, so the rates describe a fleet thinning over half an hour. 'Delay' is the link setting, not the measured age of what craft hear.
research/design/s5_lat_fresh.md:53 (Caveats on the sentence)
What changed
Nothing.
Numbers and source references
- 84
- half-hour runsresearch/design/s5_lat_fresh.md:53
- 0.05
- ordinary craft lost per craft-hour with perfect informationresearch/design/s5_lat_fresh.md:53
- 0.10 / 0.16 / 0.25 / 0.47 / 0.71
- lost per craft-hour at 0.25 / 0.5 / 1 / 2 / 3 s delayresearch/design/s5_lat_fresh.md:53 (Map sentence)
- +0.22 per second
- slope of a straight-line fit, one layout setresearch/design/s5_lat_fresh.md:53
- +24 %
- rise in time inside the bubble from perfect information to 3 sresearch/design/s5_lat_fresh.md:53 (Descriptive)
- 2.8×
- rise in deep time in the bubble, perfect information to 3 sresearch/design/s5_lat_fresh.md:53 (Descriptive)
- 0.25 % → 3.7 %
- share of craft's continuous periods inside a bubble (possibly spanning several partners) that ended in a loss, perfect information to 3 sresearch/design/s5_lat_fresh.md:53
- 600
- craft at the start of each runresearch/design/s5_lat_fresh.md:53 (Map sentence)
- six
- new layoutsresearch/design/s5_lat_fresh.md:53
- 5 updates per second
- update rate tested at 0.25 s delayresearch/design/s5_lat_fresh.md:53
- 0.25–3 s
- range of delays testedresearch/design/s5_lat_fresh.md:53
- 0.65× and 1.90×
- 0.5 s and 2 s losses relative to 1 s on these layoutsresearch/design/s5_lat_fresh.md:53
- 0.63× and 2.06×
- same ratios on the earlier layouts (cycle 254)research/design/s5_lat_fresh.md:53