This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
The stress tests changed many settings at once and could not say which one mattered. This round changed one condition at a time from a calm starting point.
What was tested
Fifteen families, two layouts, 600 s per run, eight conditions: crowding, speed, spacing, wind, turbulence, storm, link and vehicle. We counted collisions per 1,000 craft-km, plus deliveries and time in close calls. A combined case checked that collisions could be detected at all.
What was found then
No weather, crowding or speed setting caused a collision for an ideal craft, even at the extremes. Those reached 4× normal crowding and 30 m/s cruise speed. Wind reached 15 m/s, turbulence strength 4, storm strength 18. The combined check gave 2 collisions, so detection works. Weather cost deliveries instead: turbulence 4 halved deliveries per km. Only the link (first contact at 0.5 messages a second) and the vehicle's agility limits produced collisions.
What we learned later
Under the stricter acceptance rule added in cycles 4–5, turbulence 4 counts as degraded, not clean, despite zero collisions: its deepest entry into a safety bubble (0.57 of the way in) passed the 0.5 limit for clean.
research/envelope.md (Per axis, turbulence)
On a later lane grid the ideal craft lost 4 craft in one of three runs at 22 m/s and at 30 m/s (cycle 120; six runs each in cycle 123).
research/envelope.md (Per axis, speed); git 981e7c8
What changed
Nothing in the rules (new tools to set conditions, and rates per km, added).
Corrections retained in the record
Earlier wording: Density to four times the fleet, cruise to 30 m/s, wind to 15 m/s, turbulence to 4, storms to 18: zero collisions in 900 cases.
Record correction: The aggregates hold 690 non-baseline cases on those five axes (870 with the separation axis and the baseline); zero collisions is right, 900 does not match the data file.
research/data/single_axis.json
Numbers and source references
- Fifteen
- familiesresearch/envelope.md@9d286de (What is being measured)
- two seeds
- seed rollsresearch/envelope.md@9d286de
- 600 s
- run lengthresearch/envelope.md@9d286de
- eight
- axesresearch/envelope.md@9d286de
- 1,000 craft-km
- rate denominatorresearch/envelope.md@9d286de
- 4×
- densityresearch/data/single_axis.json
- 30 m/s
- speedresearch/data/single_axis.json
- 15 m/s
- windresearch/data/single_axis.json
- 4
- turbulenceresearch/data/single_axis.json
- 18
- stormresearch/data/single_axis.json
- 2
- collisions in the detector checkresearch/envelope.md@9d286de (Is the harness even able to fail?)
- 0.5 Hz
- link rung with the first contactresearch/data/single_axis.json