This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Nearly all remaining collisions were at the landing pads, where hundreds of craft converge on one pad.
What was tested
Three rounds of pad fixes, each checked on the full test set (every scenario family and variant, about 400 cases).
What was found then
The push away from the ground fought the landing. Clearance was measured in the wrong space, and the holding circle sat too close. Waiting craft were speed-matched into a brake, and braking ignored speed. Craft joined their approach lane too late, the wrong lane was cleared, and released craft flew through waiting ones. Over three rounds, collisions fell from 329 to 33. Releasing the waiting craft nearest the pad first took the one-at-a-time pad family from 44 to 10.
What changed
A ground floor at pad height for landing craft; clearance measured in 3-D; holding circle three required spacings out; waiting craft exempt from speed-matching; zigzag waiting spots; braking capped by speed; craft join the lane at release; clearance for the soonest landing; release nearest-first.
Numbers and source references
- 329 → 158 → 61 → 33
- full-matrix collisions after runs 7, 8, 9 and 10README.md (Stress campaign); git 66ef6de, 846c17d, b085ceb
- 44 to 10
- metered pad family with inside-out releaseREADME.md (Stress campaign); git 846c17d