This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
What pushes the fast craft sideways, off its line?
What was tested
Steering records on eight 3× runs, the route reversed, and the controller's small built-in clockwise turn on every dodge switched off for comparison.
What was found then
The craft's own dodges asked for a north-west push almost as large as the pull back toward its lane. In these requests, before aircraft limits, about two thirds came from the built-in clockwise turn. Ignoring speed-matching roughly doubled the dodging, and the push grew 3.7 times. Reversing the route flipped the emergency craft's side; a fast craft without priority stayed downwind. With the turn off, every setup shifted right (90–570 m) and ordinary losses fell everywhere, but emergency-craft losses moved to another edge. The turn was read as deciding which way a chain of dodges grows. Two layouts per case.
What we learned later
The second-by-second reconstruction recorded steering requests before aircraft limits and wind, not the motion actually flown; 'the turn decides what follows' is a guess from two passes.
research/reviews/auditor-c221.md
Still a candidate cause, on reused layouts.
research/design/prio_stage4_close.md (Established 3)
What changed
nothing (the turn switch was tested for comparison only; counters added)
Corrections retained in the record
Earlier wording: every craft now sits south-east of its line ... The drift is always to the left of the track, whichever way it points
Record correction: The cycle's own table shows only the package craft flipped world side; the plain and flocking-restored craft stayed downwind on both headings.
research/design/prio_falsify_record.md (lede corrected c218)
Numbers and source references
- eight
- instrumented 3× runsresearch/design/prio_decomp_record.md
- two thirds
- share of the net north-west request from the clockwise twistresearch/design/prio_dodge_side_record.md
- 3.7×
- push multiplier from the package ignoring flockingresearch/design/prio_dodge_side_record.md
- −174 → +203 m
- package offset, route reversedresearch/design/prio_falsify_record.md
- +33 → +163 m
- plain craft offset, route reversed (stays downwind)research/design/prio_falsify_record.md
- 90–570 m
- rightward shift of every tested setup with the twist offresearch/design/prio_bias_off_record.md
- 16→10, 19→10, 21→14, 32→20
- ordinary lost per two runs, twist on → offresearch/design/prio_bias_off_record.md (B3)
- 6 / 8
- runs for the reversed route and the twist-off comparisonresearch/design/prio_falsify_record.md; research/design/prio_bias_off_record.md