This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Slow-responding craft keep extra room in their safety bubble, growing with cruise speed, yet the wing's losses still rose from 10 to 95 per run. Idea: drop that room and react earlier, treating a closing neighbour as nearer than predicted by how far the pair closes during the craft's response delay.
What was tested
39 calm crossing runs (600 wing craft, 30 minutes, three layouts each) against the old controller and controls, one with the room simply removed. A new meter re-flew 27 runs to time pairs inside the old bubble.
What was found then
Losses fell to about a third at the higher speeds (29 instead of 95 at 22 m/s), with the idealised instant-response craft unchanged. They still grew with speed, less steeply (5.5× instead of 9.5×). The cost: pairs entered the old bubble about twice as often and long stays lasted 1.5–2.9× longer, though deep approaches (over halfway in) were no more frequent. Removing the room alone was worse at every speed.
What we learned later
A later version with no extra room and stronger early reaction (on steering only) failed the storm test badly (wing collisions 308 to 750), so dropping the extra room was abandoned.
research/data/oos/tau_adopt.md
The researcher noted that the crossing version tried in c125–c131 always combined early reaction with removing the extra room; the effect of early reaction alone was only separated at c133.
research/reviews/researcher-c134.md item 1
What changed
nothing (both new switches off by default)
Corrections retained in the record
Earlier wording: removing the bubble without the charge lost 15 / 35 / 53 — so the charge, not the bubble's removal, carries the gain
Record correction: The cycle's own readings show that removing the bubble alone already beat today's controller at 22 m/s (53 vs 95, classed BUBBLE-OFF-BETTER) and was flat at 16. The charge then added a further cut at every speed (CHARGE-BETTER).
research/data/oos/tau_avoid.md (Registered readings R1)
Numbers and source references
- 10 / 51 / 95
- wing craft lost per run under today's controller at 8 / 16 / 22 m/s (10 / 51.3 / 95.3)research/data/oos/tau_avoid.md (R1)
- 39
- new runs in the pilotresearch/reviews/brief-c126.md:37
- 600
- craft per runresearch/data/oos/tau_avoid.md
- 30 minutes
- run length (1 800 s)research/data/oos/tau_avoid.md
- two
- bigger-bubble controls (bubble coefficient 2 and 3)research/data/oos/tau_avoid.md (R2)
- 27
- runs re-flown with the per-pair meterresearch/data/oos/refcensus_pair.md
- 5 / 15 / 29
- wing craft lost per run with the charge and no bubble (5.3 / 14.7 / 29.3)research/data/oos/tau_avoid.md (R1)
- 5.5× vs 9.5×
- 22 m/s loss / 8 m/s loss, new rule vs today'sresearch/data/oos/tau_avoid.md (R6)
- about twice
- episodes inside the old spacing per craft-hour, new vs today's (2.57 / 1.87 / 1.88×)research/data/oos/refcensus_pair.md (Registered readings)
- 1.5–2.9×
- ratio of averaged run-level 90th-percentile episode durations, new vs today's (1.53 / 2.05 / 2.88)research/data/oos/refcensus_pair.md (Registered readings)
- 0.85 to 0.54
- deep (at least half-depth) episodes per craft-hour at 16 m/s, today's vs newresearch/data/oos/refcensus_pair.md (Registered readings)
- 163 to 53
- episodes reaching three-quarters depth at 22 m/s, three seedsresearch/data/oos/refcensus_pair.md (table, depth bins)