This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
The 3× losses happened at the west edge, where craft reappear at their lane start. Would moving the route start help, and why does the craft drift there?
What was tested
Moved the route start 150 m inward, traced the drift, and gave back one rule at a time; one craft also flew alone, 200 m off its line.
What was found then
Moving the start did not help (4 losses either way). The craft ended trips north-west of its line, from 16 m at 1.5× to 174 m at 3×, carrying over between trips. The corridor did not cause it and was not needed at 3×. Giving back speed-matching brought it to about 30 m and cut losses from 4 to 1 in 10 runs, but its slow trips returned. Flying alone, it came back within one trip. Read then: at 3× the crashes are the price of speed.
What we learned later
'The price of speed' was marked a guess, not a proven trade-off.
research/reviews/auditor-c215.md
A candidate cause, on reused layouts. One switch changed speed, timing, offset and survival together, so the chain from offset to loss is not isolated.
research/design/prio_stage4_close.md (Established 3)
What changed
nothing
Numbers and source references
- 150 m
- route start moved inside the gridresearch/design/prio_trackend_record.md
- 20 / 30 / 30
- runsresearch/design/prio_trackend_record.md; research/design/prio_wander_record.md; research/design/prio_term_record.md
- 4 vs 4
- 3× losses with corner start vs moved startresearch/design/prio_trackend_record.md
- 16 / 74 / 174 m
- package median lap-end offset north-west at 1.5× / 2× / 3×research/design/prio_wander_record.md
- +29 / +30 m
- median lap-end offset with flocking restoredresearch/design/prio_term_record.md
- 1 vs 4 in 10
- 3× losses with flocking restored vs packageresearch/design/prio_term_record.md
- 2.12–2.37 vs 1.40–1.64
- 90th-percentile lap ratio, flocking restored vs packageresearch/design/prio_term_record.md
- 670–920 m
- distance over which a 200 m offset fell to about 37 % in an empty skyresearch/design/prio_term_record.md (Step 2)
- one lap
- time for the lane pull to bring it back in an empty skyresearch/design/prio_term_record.md (Step 2)