This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Early reaction (steering away sooner) with full extra room had improved every test set flown. Did it pass the full adoption rule, and were its remaining close calls deep or shallow?
What was tested
The agreed rule: the idealised instant-response craft unchanged, no fixed test set (storm or standard mixed) clearly worse, and crossing losses at least halved for every slow airframe at every speed. A new gauge timed pairs more than halfway into the old bubble, on the wing and 27 runs of other airframes. The adopt-or-hold question went to the researcher.
What was found then
The test-set half passed: the instant-response craft was identical and every set was better or unclear. The crossing half failed: only 9 of 12 conditions halved losses, and losses still grew with speed on every slow airframe. At 22 m/s, deep time inside the old bubble fell to between a third and half the old level, though each deep stay lasted as long or longer. A quarter-size room cut losses to a quarter but kept 86 % of the old deep time.
What we learned later
A three-quarter room with stronger early reaction later passed the full rule, so the question to the researcher became moot.
research/data/oos/tau_j2gate.md
What changed
nothing
Corrections retained in the record
Earlier wording: beats it on every frozen test set for every slow-reacting type
Record correction: The multirotor and eVTOL matrix results were inconclusive (too few collisions), and the delayed-sideways craft had no benchmark run. Reviewers struck 'better everywhere flown'; the correct reading is 'better or inconclusive, none worse'.
research/data/oos/tau_gate2.md; research/reviews/auditor-c138.md (F-TAU-GATE-SCOPE)
Earlier wording: the losses still grow with cruise speed on every slow type, which is the one thing this stage was opened to fix. So the mechanical adoption rule says no
Record correction: The rule's crossing clause failed because 3 of 12 conditions did not halve losses (PARTIAL). Growth with speed is a separate, within-setting test that reviewers said had been conflated with it.
research/data/oos/tau_gate2.md; research/reviews/auditor-c136.md (F-TAU-GATE-SPEED)
Numbers and source references
- six
- settings on the wing time-at-depth readresearch/data/oos/tau_sev.md
- 27
- runs for the three other airframes at 22 m/sresearch/data/oos/tau_sev2.md
- 308 to 217; 544 to 158; 67 to 48; 84 to 49
- collisions, today's vs full-bubble charge: wing storm, wing matrix, multirotor storm, eVTOL stormresearch/data/oos/tau_gate2.md
- 9 of 12
- crossing conditions that halved losses (PARTIAL; the rule needed all 12)research/data/oos/tau_gate2.md (Stage reading)
- 0.50 / 0.54 / 0.55 / 0.33
- deep exposure vs today's at 22 m/s: wing / eVTOL / delayed-sideways / multirotor (wing 1.19/2.39)research/data/oos/tau_sev.md; research/data/oos/tau_sev2.md
- 86 %
- quarter bubble's deep exposure as a share of today's at 22 m/s (2.06 / 2.39)research/data/oos/tau_sev.md
- a quarter
- quarter bubble's losses vs today's at 22 m/s (23.3 vs 95.3)research/data/oos/tau_sev.md