This is an earlier experiment. Read the amendments below before treating its findings as current.

The question

A few very deep close calls kept the wing tests from counting as safe, and the tool locating them was wrong.

What was tested

We fixed the recorder. In this test, a craft reaching its lane end is put back at the start. We tried two ways: re-centre it on the lane, or wait until the spot is clear (144 runs).

What was found then

All eight very deep close calls were re-entries: a craft put back at the lane start kept its sideways drift and landed on another craft. Re-entries are under 2 % of collisions. Re-centring on the lane is badly unsafe (about 6–8 times the collisions). Waiting until the spot is clear and placing the craft on the start line gave no deep close calls in 48 runs. So these events were a flaw of the test setup, not the controller failing.

What we learned later

LATER / CYCLE 40

'At no cost' for waiting was taken back: it was not shown to be no worse, and deep close calls lasted longer.

research/reviews/brief-c41.md

LATER / CYCLE 41

The clear-spot placement raised losses in one case in 11 of 12 paired layouts (about 22 more craft per run).

research/reviews/brief-c42.md

What changed

Nothing. The re-entry methods stay test settings; the recorder fix is in the simulator.

Corrections retained in the record

  • Earlier wording: Putting each re-entering craft back on its lane’s centreline is badly unsafe — seven to eight times the collisions

    Record correction: the cycle's own counts give 5.7–7.6× (the brief also wrote ×7–8)

    research/data/oos/reentry.md

Numbers and source references
29 of 48
runs where the old recorder named the wrong craftresearch/reviews/brief-c39.md
24 + 24
re-run and new runs after the fixresearch/reviews/brief-c39.md
144 / 96
runs in the re-entry and tagging testsresearch/reviews/brief-c40.md
12 of 710
collisions that were lane re-entriesresearch/reviews/brief-c39.md
5.7–7.6×
collisions with re-centring (114→861, 144→971, 212→1 234, 240→1 378)research/data/oos/reentry.md
8 of 8
very deep near-misses that were re-entriesresearch/reviews/brief-c41.md
0 of 48
runs with a very deep near-miss under the fixed holdresearch/reviews/brief-c41.md
Original structured record ↗