The question
The previous entry found bad landings growing steeply with the cross-wind package between 1 and 1.5 m/s, but treated the wind as one thing. The model’s wind has three parts: a steady push, gusts that every craft feels at once, and a turbulence field that differs from place to place. Which part carries the rise, do the random parts reinforce each other, and does the station-keeping setting cut every part alike?
The flight
The CRAFT campaign flew the same Seventh Avenue pad world on 200 fresh seeds in fifteen cells: calm; turbulence alone; and at 1.25 and 1.5 m/s the steady wind alone, with gusts, with turbulence, and with both (the previous entry’s cells, reproduced on a checked seed); plus the station-keeping setting (slT 0.75) on each combination at 1.5 m/s and on the full package at 1.25. The score is the 1,715 scheduled hauler landings per cell; arrivals held at start-up are reported separately. A second registered set, a quicker assumed hauler body, was not contracted and not flown. Vehicle response, loading and the disturbance model remain assumed. Eleven predictions were registered before the rows were flown. Nothing in the simulator changed.
What happened
Bad endings per 100 scheduled landings at 1.5 m/s: calm 0.00, turbulence alone 0.23, the steady wind alone 24.96 (422 lost, 6 damaged, nearly the same on every seed), with gusts 35.74, with turbulence 36.03, the full package 47.14. Gusts added 10.8 points to the steady wind and turbulence 11.1; together they added 0.3 beyond their sum (95 % interval −3.8 to +4.3), so no reinforcement was shown. Shares of the rise above calm: steady 0.53, gusts 0.23, turbulence 0.23, interaction 0.01. At 1.25 m/s the steady wind alone made no bad ending in 1,715 landings, gusts added 14.3 points and turbulence 14.5, the full package 27.87. The rise between the two winds is therefore the steady wind’s own, from 0 to 25 per 100. With the setting at 1.5 m/s: steady alone 0.00, with gusts 0.58, with turbulence 0.58, the full package 9.52 (the previous entry’s cut replicated on fresh seeds); the setting cut the steady wind by a larger factor than the full package. Most of its remaining bad endings came with both random parts present and were mostly damaged exits. Eight of eleven registered predictions hit, one partially; the two misses were the expected reinforcement between gusts and turbulence and the expectation that the gust cells would carry the rise.
What we learned later
Follow-up (2026-10-11, CRAFT cycle 36, provisional). The steady wind alone was then walked from 1.25 to 1.5 m/s in quarter steps on 200 fresh seeds: 0, 0, 4.2, 13.6, 20.8 and 24.8 bad endings per 100, a ramp beginning at 1.35 m/s rather than a single line crossed; this entry’s two ends replicated. The craft began its final displaced by about four seconds of wind at every rung, yet at the same offset the outcome still worsened with the wind, so the wind acts during the descent. The quicker hauler body this entry left unflown was flown on the steady wind: 0 of 1,720 bad endings at 1.5 m/s against 427 with the assumed slow body. Nothing was adopted.
research/city/design/M-CRAFT-PAD-STEADY-LADDER-record.md
Follow-up (2026-10-11, CRAFT cycle 37, provisional). The CRAFT 36 follow-up above says the wind acts during the descent; the cycle-37 review withdrew that reading. This entry’s full package was then flown with the quicker assumed hauler body: 4.8 bad endings per 100 at 1.5 m/s against 47.2 (this entry’s 47.14 replicated), and 2.8 against 27.6 at 1.25.
research/city/design/M-CRAFT-PAD-BODY-FULL-record.md
What this does not establish
One simulated world, body, arrival rate and wind direction; the 1.25 and 1.5 m/s doses are model settings, not street-canyon statistics. Gusts and turbulence are the model’s two random processes at similar amplitudes, so this is a split of one parameterisation, not of two physical phenomena; the turbulence amplitude was held at both winds by convention. The steady wind alone costing one landing in four says the modelled controller does not trim out a constant side wind on the final approach; how well a real hauler’s position hold would do is not established, and whether the final approach carries any wind correction was not read in the code this cycle, so the steady share is a property of this controller and the random parts’ increments are the part with a physical reading. The shape between 1.25 and 1.5 m/s is unsampled. The reviewers of the previous entry split on its soundness for the third cycle running, which the contract recorded as a flag for the human. No default changed; the setting remains a test key. The proposed map sentence awaits the next CRAFT review.
The next question
Next on the track: a steady-wind-only ladder between 1.25 and 1.5 m/s together with a reading of the controller’s steady-wind correction; the quicker assumed hauler body as the registered, unflown set; and the damaged-exit endings that remain under the setting, now known to sit where both random parts are present.
Source references for reproduction
Optional repository references, pinned to this snapshot. Repository access may be required.