The question
The previous entry found that the steady part of the wind alone spoiled no landing at 1.25 m/s but one in four at 1.5 m/s. Where between those winds does the loss begin, is it simply that the craft is blown further sideways before its final descent, and how much of it is the slow response assumed for the hauler (1.5 s to respond, 0.3 s delay)? The two main reviewers disagreed for the fourth cycle running on which question came first, so both were registered with sixteen predictions and flown together in one job on the same seeds.
The flight
The CRAFT campaign flew the Seventh Avenue pad world on 200 fresh seeds with the wind reduced to a steady side push, no gusts and no turbulence, which is a model dose rather than a street-canyon measurement. Eleven cells: the push at 1.25, 1.30, 1.35, 1.40, 1.45 and 1.50 m/s; the station-keeping setting (slT 0.75) at 1.40 and 1.50; and the hauler given the heavylift class’s quicker response (0.6 s and 0.15 s) at 1.25 and 1.50, with and without the setting. The score is the 1,720 scheduled hauler landings per cell; arrivals held at start-up are reported separately. The simulator itself was not changed.
What happened
Bad endings per 100 scheduled landings: 0 at 1.25 and 1.30 m/s, then 4.2, 13.6, 20.8 and 24.8 at 1.35, 1.40, 1.45 and 1.50 (427 of 1,720 at the top, nearly all of them losses). No single quarter-step carried half of the rise, so the registered reading is a ramp, not a step; the plan had predicted a later, sharper onset. The previous entry’s two ends replicated.
The slow-bodied craft began its final approach displaced sideways by about four seconds’ worth of wind at every rung (5.4 m at 1.25, 6.5 m at 1.5). But at the same starting offset the outcome still depended on the wind: in the 5 to 6 m group, none went bad at 1.25 m/s and 46 per 100 at 1.5. The plan had predicted one shared curve by offset; that missed. The offset is a symptom, and the wind keeps acting during the descent.
With the station-keeping setting, 0 of 1,720 at 1.40 and at 1.50 m/s. With the quicker response, 0 of 1,720 at 1.25 and at 1.50, with or without the setting, and the final began less than a metre to the side. All seven quicker-body predictions held; the ladder scored six hits, one partial and two misses.
What we learned later
Follow-up (2026-10-11, CRAFT cycle 37, provisional). The cycle-37 review corrected this entry’s record: the bins at equal starting offset disagree, but the rows cannot time that disagreement, so ‘the wind keeps acting during the descent’ is withdrawn; most losses came at the first wall contact high in the final. The quicker body’s cut moves with the response together with the planning the simulator ties to it. Separated on fresh seeds under the steady wind alone, the quick machine with the slow planning made 0.5 bad endings per 100 and the slow machine with the quick planning none, so the cut belongs to neither half alone. Under the full package with gusts and turbulence the quicker body made 4.8 per 100 at 1.5 m/s against 47.2 for the slow body. Both bodies are assumed; nothing was adopted.
research/city/design/M-CRAFT-PAD-BODY-FULL-record.md
What this does not establish
One simulated world, wind direction and arrival rate, and a steady wind with no fluctuation. The hauler’s 1.5 s / 0.3 s response, its lateral acceleration limit and the wall and body distances are assumed or configured, so no real wind limit or landing risk follows, and 1.35 m/s is not the pad’s wind limit: with gusts and turbulence, landings go bad at far lower winds. The quicker-body cells change the response together with the planning the simulator ties to it, so they do not say which of the two does the work, and the quicker body was not flown with gusts or turbulence. Among arrivals held at start-up, a simulator artefact outside the score, the quicker body did worse (84 against 43 bad of 280). Reviewers of the previous entry split (one flawed, one sound); this entry has not yet been reviewed.
The next question
Nothing is adopted: the station-keeping setting stays a test key and the quicker body was a test cell, not new hauler numbers. The doubt has moved to an input: the wind-dose, wind-split and this ladder all rest on the assumed response, and finding a sourced response for a six-tonne multirotor is now an open question for the programme. Next candidates: the quicker body under the full wind package, a reading of what happens during the descent, and why the setting’s remaining endings under the full package are damaged exits.
Source references for reproduction
Optional repository references, pinned to this snapshot. Repository access may be required.