Recorded finding
Agility changes the outcome.
Slow response and weak acceleration substantially worsen the simulated loss rate. Ideal craft hide limitations that become visible when response lag is introduced.
Read the record ↗EVIDENCE REGISTER
What survived
the experiment?
Read the finding with its conditions. An integrated experiment can still have a provisional interpretation.
Boundary unresolved
Fast passage is only part of priority.
The priority programme demonstrated near-straight transit in selected conditions. It did not establish a safe operating boundary for the emergency craft.
Read the record ↗Provisional
A lower loss count can hide a new hazard.
Under own-height error, altitude assignment can exchange wall losses for contacts between craft. CITY 35 traced nearly all of them to grants made while the pair was already truly inside its safety distance; withholding those grants with perfect knowledge removed the excess on the weaker link and most of it on the better one, keeping a smaller loss reduction. A guard built only from the craft’s own estimates (CITY 36) reproduced the oracle’s result on the weakest link in calm air with a 6 m allowance (contacts 188 to 10 against ordinary handling’s 6, still 107 fewer losses, 7 per cent of grants refused), but on the better link and in a descent it trades contacts against losses; nothing is adopted.
Read the record ↗Recorded finding
An uncertainty margin also needs space.
A separation policy that activates more avoidance can produce wall losses in constrained geometry. Communication and available manoeuvring space must be examined together.
Read the record ↗Provisional
Landing depends on the whole response.
Disturbance and controller timing interact at the street-side pad. In one tested world, slowing the final approach made bad endings more frequent. Bad endings rose steeply with cross wind, from about 10 to 46 per 100 between 1 and 1.5 m/s; taken apart, the steady wind alone carried about half of the rise at 1.5 m/s and none at 1.25, with gusts and turbulence adding about a quarter each. The station-keeping setting kept its cut at every dose; with it the steady wind alone made no bad ending in 1,715 landings. Under the steady wind alone the loss begins as a ramp from 1.35 m/s, and a quicker hauler response removed it at 1.5 m/s; either the quicker machine or the quicker autopilot planning alone was enough. With gusts and turbulence the quicker body cut bad endings from 47 to about 5 per 100; quick planning alone on the slow machine halved them (47 to 23.5), the planning and the body adding up at roughly 57 and 43 per cent. More held arrivals ended badly with the stopping-only planning when the model released them from a standstill; released like other arrivals, the same planning cut their bad endings from 47 to 27 per 100, the release and the planning acting together. On eleven chosen held arrivals the release speed carried that difference (21 of 22 cases); release timing changed one ending and the pad check none. Both bodies and the planning values are assumed.
Read the record ↗Provisional
Speed-matching, not the avoidance push, carried refused craft into walls.
In the tested crossing, alignment with the other level carried the lost craft across the lane in its last seconds. Switching that alignment off removed the wall losses at a 4 m allowance, at the price of more band exits; throughput was not measured. A later cycle replicated this on fresh layouts and found that clipping the across-lane part of a refused pair’s own dodge removed most of the remaining losses, while clipping its vertical escape made them far worse. CITY 36 saw the same mechanism on its crossing under a 4 m own-height error: ordinary handling lost 36 per cent fewer craft on the weakest link and 90 per cent fewer on the better one with contacts unchanged; still not adopted. CITY 37 found the saving without height error, on wider spacing and in the demonstration fleet (91 to 4 lost), and measured its price: about 4 to 6 per cent more band exits per second of flying, and in one setting about a third more time outside the band; the default is an open decision. CITY 38 instrumented the lost craft: the term was the largest sideways push for about 9 in 10 wall losses on the better tier and about half on the worst (where avoidance of the other level is almost as common), pushing them 20 to 35 m sideways in their last 10 seconds against 1 to 6 m for their neighbours; an attribution along the flown paths, not a proven cause. CITY 39 re-flew 1,530 of those losses with the push taken away from the lost craft alone: about half would have missed the wall on the worst tier and nearly all on the better one, against 2 to 16 per cent for a forward or backward push of the same size, so the direction of the push decides; on the better tier each save cost 0.13 to 0.24 other craft lost. CITY 40 replicated the saves on new layouts and found that removing the push 30 seconds before the hit saves no more; that short-window cost on the better tier belongs to the way CITY 39 scored the removal, but over a full run every way of removing it was followed by about 0.1 to 0.5 more losses elsewhere per saved craft. LINK 31 replicated the gate hold that removes the remaining losses in calm air on new layouts (148 to 74 and 76 to 3 lost) and found it untestable in the standard wind, which removed every craft with or without it. LINK 32 (exploratory) found that craft spend about 80 per cent of their time outside their band, yet without the hold the building hits are far more likely per second inside it; the hold removes mostly those in-band losses. LINK 33 compared craft at the same place: about half of that gap was where in-band craft fly, but the rest stays, highest in the upper half of the band, with the sideways push coming from the reaction to craft in other bands; telling craft the true sideways position of those craft did not help. LINK 34 withdrew the upper-half part: almost all such losses are on one band, which mostly flies in its upper half, and compared like with like no difference between the halves was detected (LINK 35: not proof of equal rates); the sideways push comes about 60/40 from crossing-street craft above and same-street craft below. LINK 35 re-stacked the same four lanes over the same heights: the in-band losses moved to a different lane each time, so that one band is a property of the plan’s stacking, and one stacking (the avenue levels at the top and bottom, 45th Street’s two levels in the middle) lost almost no craft; why is untested, a clue rather than a recommendation. CITY 41 (2026-10-13, provisional) narrows this headline’s ‘not the avoidance push’ for the crossing’s worst tier with speed-matching on: re-flying each lost craft without its own sideways dodge away from other-level craft saved about four in ten of its wall losses, and without both that dodge and speed-matching about 96 in 100; the craft knew their heights exactly, so this is not a licence to ignore the other level.
Read the record ↗These are selected findings, not the entire evidence base.
The full map includes historical lines, withdrawn interpretations and the exact conditions of each experiment.