This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
The exact-replay checks (short runs that must reproduce to the last digit after any code change) covered only the idealised instant-response craft, leaving this stage's code unprotected. the researcher also still had to decide the wing.
What was tested
Eight 40-second test cases were built, six with slow-responding craft. Asked if anything was left worth improving, both reviewers said no, and the loop stopped itself for the first time (cycle 184). the researcher then said 'Wing', and the wing switched after reviewer checks.
What was found then
The new checks reproduce exactly, apart from tiny rounding differences in two running totals. After the switch only the wing's case changed: its live safety bubble radius grew from 43.8 to 47.8 m. The new default reproduced the tested full-room runs exactly, for example 6 crashes on the 16 m/s crossing. Untested wing variants stayed at three-quarters, guarded by an automatic check.
What changed
Wing: full extra room (c185). Untested wing variants stay at three-quarters. Exact-replay checks for slow-responding airframes added (c183).
Corrections retained in the record
Earlier wording: reference radius 43.8 m → 47.8 m
Record correction: The value that moved is the live spacing radius (rh); the forced reference radius was already 47.82 m.
research/design/tau_lagged_anchors_record.md (Baseline); research/design/tau_wing_lagk_record.md
Earlier wording: the wing's stall-speed variants inherit from the wing preset by copying
Record correction: The stall variant already kept an explicit 0.75. The variant copied from the wing preset (wing3) was the one that needed an explicit 0.75.
research/design/tau_wing_lagk_record.md (Change)
Numbers and source references
- Eight; six; two
- test cases; slow-craft cases; instant-response controlsresearch/design/tau_lagged_anchors_record.md
- 40
- seconds per test caseresearch/design/tau_lagged_anchors_record.md
- 184
- cycle the loop stopped itselfresearch/design/tau_bubble_stage_close.md
- 43.8 to 47.8 m
- wing live spacing radius (rh) before and after the switch; the reference radius stayed 47.82 mresearch/design/tau_wing_lagk_record.md; research/design/tau_lagged_anchors_record.md
- 6
- crashes, wing 16 m/s crossing, new default = tested full-bubble runresearch/design/tau_wing_lagk_record.md
- 166 / 321
- wing storm-set crashes / craft lost, new default = tested runresearch/design/tau_wing_lagk_record.md