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

The question

the researcher had delegated settings changes to reviewer agreement, so four keep-or-switch questions were settled inside the research loop by a rule fixed before scoring.

What was tested

Keep the current default if the full extra room lost more craft or collided more on the storm or mixed test set, spent more time close in storms, had a cruise-speed cost that repeated on fresh layouts, or couldn't be scored. Switch only if it lost fewer craft, and no more on at least three of five layouts.

What was found then

The rule said switch the multirotor, slow-responding test craft and wing; the eVTOL keeps three-quarters, as its full room lost slightly more craft on both test sets. Reviewers held the multirotor, whose win rested on one layout, and split on the wing. The test craft switched at cycle 181. On five fresh layouts the multirotor's full room lost 329 craft against 401, no worse on 4 of 5, so it switched at cycle 182. Each switch reproduced its tested runs exactly; the idealised instant-response craft was unchanged.

What we learned later

LATER / CYCLE 185

The wing, left to the researcher after the reviewers split, switched to the full extra room on his decision.

research/design/tau_wing_lagk_record.md

What changed

Extra room now set per airframe: full for the slow-responding test craft (c181) and multirotor (c182); the eVTOL and wing stay at three-quarters for now.

Numbers and source references
three of five
offsets on which the full bubble must lose no more craft (ties allowed) to switchresearch/design/tau_bubble_decision_rule.md clauses 5–6
329 vs 401
multirotor storm-set craft lost on fresh seeds 500–900, full bubble vs adoptedresearch/data/oos/mr_c5_gate_hold.md; research/design/tau_mr_lagk_record.md
4 of 5
fresh seeds where the full bubble was no worseresearch/data/oos/mr_hold_prereg.md
181; 182
switch cycles for the delayed-sideways craft and multirotorresearch/design/tau_lagk_preset_record.md; research/design/tau_mr_lagk_record.md
Original structured record ↗