This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Last stage, ordinary traffic lost dozens of craft per half hour, worst in calm air with a 1 s link delay, on just two layouts. Is it real, and which part of the link causes it?
What was tested
36 half-hour runs on six fresh layouts compared the link, perfect information and wind. More runs dropped the emergency craft or changed one link setting at a time.
What was found then
Calm with the 1 s link lost 3.8× more ordinary craft than perfect information, on every layout. Without an emergency craft the gap was 5.6×, so it is the traffic's own problem. Delay matters most: 0.5 s cut losses to about two-thirds, and 2 s doubled them. More frequent updates or no lost messages fell short of the threshold fixed in advance (a 19 % drop may still be real). Time inside the bubble barely moved.
What we learned later
'Update rate barely matters' was softened to 'did not pass the threshold': more updates per second do make the picture each craft hears fresher; they just did not move losses enough.
research/design/s5_link_dose.md:49 (Reading, c255 note)
The 'age of the heard picture' figures used here came from a steering setting, not from measuring how old messages really were; lost messages make the real age longer.
research/design/s5_link_dose.md:49 (Reading, c256 note)
What changed
Nothing.
Corrections retained in the record
Earlier wording: First test (it changes no settings): 36 half-hour runs on new seeds.
Record correction: 24 new runs plus the 12 calm 1 s runs already flown in cycle 252 on the same fresh seeds 200–700.
research/design/s5_proposal.md (First flight, Rows) and :56
Numbers and source references
- 36
- half-hour runs in the first test (24 new, 12 from cycle 252)research/design/s5_proposal.md:56
- 3.8×
- calm 1 s link vs calm perfect information, ordinary craft lost per craft-hourresearch/design/s5_proposal.md:56
- 0.246 vs 0.065
- ordinary craft lost per craft-hour, calm with 1 s link vs perfect informationresearch/design/s5_proposal.md:56
- 2.4×
- calm 1 s link vs 5 m/s wind with the same linkresearch/design/s5_proposal.md:56
- 12
- runs without any emergency craftresearch/design/s5_link_dose.md (Spec, Rows A)
- 5.6×
- link vs perfect information with no emergency craft (0.254 vs 0.045)research/design/s5_link_dose.md:49
- 60
- runs changing one link setting at a timeresearch/design/s5_link_dose.md (Spec, Rows B)
- 0.60–0.66×
- losses at 0.5 s delay relative to 1 sresearch/design/s5_link_dose.md:49
- 2.06×
- losses at 2 s delay relative to 1 sresearch/design/s5_link_dose.md:49
- 0.81–0.97×
- losses with 2 or 5 updates per second, or no lost messages, relative to the 1 s bundleresearch/design/s5_link_dose.md:49
- 19 %
- possible real drop with no lost messages (plain arm 0.81×) that did not clear the barresearch/design/s5_link_dose.md:49 (Reading)
- 53–60 s per hour
- time inside the bubble across delay settingsresearch/design/s5_link_dose.md:49 (Reading)
- six
- fresh layouts (the cycle-252 layouts, reused for the calm 1 s runs)research/design/s5_proposal.md (First flight, Rows)