This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
A collision rate without knowing how collisions happen explains nothing. The simulator now records the true colliding pair and its speed over the ground at contact.
What was tested
We read back 482 collisions from the cases with five or more.
What was found then
Two thirds were crossings (67 %); the rest were head-on (20 %) and overtakes (13 %). Most (61 %) happened away from the pads, with a typical height difference of 2.4 m. Two families carry 73 %. One lane crossing alone, a fast express lane into an eastbound lane 20 m below, carries 32 %. Crowding fails on the way; the link fails near the pads. The first draft called the eastbound lane uncooperative. In fact its craft are never removed on contact, so only the other side's losses count. Four runs to tell causes apart were queued.
What we learned later
The four runs (2026-09-11): a slower express lane and a doubled height gap changed nothing clearly; keeping the eastbound lane at its design number of craft cut the count to a third (111 → 36). The slow-responding craft's lanes were found sinking toward the ground, and holding them up did not cut the count.
research/envelope.md (What the collisions are); research/reviews/researcher-c5.md (Addendum)
The sinking was traced to slow-responding craft descending more easily than they climb (cycles 6–7); that explains the sinking but was not shown to cause the collisions in the crowded case.
research/reviews/researcher-c6.md, researcher-c7.md; research/envelope.md (What would falsify this)
What changed
Nothing in the rules (the simulator now records the true pair and its ground speed).
Numbers and source references
- 482
- collisions readresearch/envelope.md (What the collisions are)
- five
- minimum collisions per cell includedresearch/envelope.md (What the collisions are)
- 67 %
- crossingsresearch/envelope.md (What the collisions are)
- 20 %
- head-onresearch/envelope.md (What the collisions are)
- 13 %
- overtakesresearch/envelope.md (What the collisions are)
- 61 %
- en-routeresearch/envelope.md (What the collisions are)
- 2.4 m
- median height difference at contactresearch/envelope.md (What the collisions are)
- 73 %
- share of two familiesresearch/envelope.md (What the collisions are)
- 32 %
- share of one lane crossingresearch/envelope.md (What the collisions are)
- 20 m
- vertical gap between the two lanesresearch/envelope.md (What the collisions are)
- 80 of 106
- that pair at lagged 4× densityresearch/envelope.md (What the collisions are)
- 4×
- density multiplierresearch/envelope.md (What the collisions are)