This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
So far we measured the dodge the steering asked for, not what the aircraft did once its slow response caught up. Do colliding pairs push apart less, and does that shortfall grow with delay?
What was tested
36 runs re-flown with a recorder that does not change behaviour, measuring how hard each pair accelerated away from each other 1–2 s after they met.
What was found then
With the link on, colliding pairs pushed apart much less than survivors (typically 0.6–0.8 against 2.0–2.4 m/s²). The shortfall did not widen with longer delay; it narrowed slightly. By its own rule the test is inconclusive: it needed at least 20 fatal pairs per group, and perfect information had 13 and 12. It had first been reported as a clear no. What puts a pair into this weak-push state was not measured.
What we learned later
That filter still let through crashes on the last step of the measurement. Restated using only fatal episodes longer than 2 s: shallow 1.80 → 1.30, crossings 2.13 → 1.50; the direction is unchanged.
research/design/s5_carrier_real.md:70 (c265 note)
The pair-by-pair search was closed. Four clues mark colliding pairs (error in the heard position, how hard they were told to dodge, meeting angle, how hard they pushed apart), but none passed its test, set in advance, for growing with delay.
research/design/s5_angle_close.md:95 (Open)
What changed
Nothing. A recording tool was added, off by default.
Numbers and source references
- 36
- runs re-flown with the hookresearch/design/s5_carrier_real.md:45
- 0.6–0.8 vs 2.0–2.4 m/s²
- median push apart, colliding vs surviving pairs, with the linkresearch/design/s5_carrier_real.md:45
- 20
- fatal pairs required per meeting type and settingresearch/design/s5_carrier_real.md:45
- 13 and 12
- fatal pairs with perfect information, shallow / crossingresearch/design/s5_carrier_real.md:45
- 1–2 s
- measurement interval after the pair meetsresearch/design/s5_carrier_real.md:45
- 1.80 → 1.26
- gap between surviving and colliding shallow pairs, 0.5 s → 2 s, full windows (cycle 264)research/design/s5_carrier_real.md:70
- 2.13 → 1.50
- same gap for crossingsresearch/design/s5_carrier_real.md:70