This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
A rule pulls straying craft back to their lane at a capped speed. Is the cap too low, and could a stronger or faster pull cut the wing's losses?
What was tested
37 new runs, ideal craft (instant response) and wing, 1–3 layouts each: cap raised to full cruising speed, pull 2–8× stronger, and a recorder for when the command hit its cap.
What was found then
Raising the cap cut the ideal craft's straggling by about 40 % at full heading copying, less at weaker copying, but cost craft: the wing lost 64–128 more per run, and the ideal craft 20 where it had lost none. Among stragglers the command sat at its cap nearly always. A 4× stronger pull trimmed straggling by a fifth. Three predictions set in advance about why all failed. On the wing, only 4× strength beat the baseline on all 3 layouts. The thread closed with no reliable lever found.
What we learned later
Withdrawn: 'the cap only matters at full copying'; the pull-back hits its cap beyond about 108 m off lane at every strength.
research/envelope.md:4742
Narrowed: a command at its cap is not a measured return speed, and 'the stronger pull costs no craft' rested on one run.
research/envelope.md:4742
Narrowed: 'no lever' became 'average effect unclear' (about 16 fewer wing losses averaged over 4–6× strength, found after the fact). A failed test does not rule the suggested causes out.
research/envelope.md:4527
What changed
Nothing. A recording switch for the pull-back was added, off by default.
Numbers and source references
- 37
- new runs (8 + 6 + 6 + 6 + 11)research/envelope.md:4527; research/envelope.md:4603; research/envelope.md:4666; research/envelope.md:4742; research/envelope.md:4791
- 2–8×
- pull-back strength range (0.1 → 0.2–0.8)research/envelope.md:4603
- 1–3
- random starts per settingresearch/envelope.md:4603
- 40 %
- tail cut from the higher cap at full copying (ratios 0.59 / 0.63)research/envelope.md:4791; research/envelope.md:4742
- 64–128
- extra wing craft lost per run under the higher capresearch/envelope.md:4791
- 0 → 20
- ideal craft lost under the higher cap at full copyingresearch/envelope.md:4791
- 1.00 / 0.96
- share of in-tail samples with the command at its cap, ideal craft / wingresearch/envelope.md:4742
- a fifth
- tail cut from 4× pull-back strength (ratio 0.79)research/envelope.md:4742
- three
- registered predictions that failedresearch/envelope.md:4527; research/envelope.md:4603; research/envelope.md:4666
- 84 / 84 / 90 vs 100 / 100 / 102
- wing craft lost at 4× strength vs baselineresearch/envelope.md:4603