This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Review of the six-layout step checked the measures the tables were built on.
What was tested
Four measurement faults were fixed, leaving collision reference runs unchanged. Close-call time was remeasured over pairs that can actually collide. We checked that collisions arrive at random, as the confidence ranges assume, and ran one layout on two processors.
What was found then
The faults: colliding pairs were left out of closeness measures, contact speed used airspeed, closing speed was not along the line between craft, and craft that cannot be lost set the worst approach. Old close-call times were 1.0–1.9 times the corrected ones, not 5 times. The two measures disagree: the slowest link is 15 times calm on collisions but 1.2 times on close-call time. The randomness check passed in 19 cases. Intel and Apple processors play the same layout differently, but swapping two layouts moved only two collisions. New rule: nothing with more losses than calm is called safe.
What changed
Four measurement fixes; close-call time counted only for pairs that can collide; an acceptance rule (clean, degraded or out).
Numbers and source references
- Four
- metric defectsresearch/envelope.md (What cycle 3 found)
- 1.0–1.9×
- exposure inflation per cellresearch/envelope.md:6196
- 5×
- inflation first suggested by one caseresearch/envelope.md:6196
- 15×
- 0.5 Hz link collision rate against calm (1.22 vs 0.08)research/envelope.md (exposure table)
- 1.2×
- 0.5 Hz link exposure against calm (2.2 vs 1.8)research/envelope.md (exposure table)
- 4×
- cruise 30 exposure against 4× density (47.6 vs 11.8)research/envelope.md (exposure table)
- 40 %
- cruise 30 collision rate relative to 4× densityresearch/envelope.md (exposure table)
- 19
- cells in the Poisson dispersion checkresearch/reviews/brief-c5.md:106
- 30
- cruise speed, m/sresearch/envelope.md (exposure table)
- two
- collisions moved by replacing two Mac seedsresearch/reviews/researcher-c5.md:95