CITY / 41
2026-10-12
On the worst assumed radio tier, a lost craft’s sideways dodge away from craft on the other height level is the second move that carries it into walls. Re-flying each wall loss with that dodge and its speed-matching both taken away saved about 96 in 100; speed-matching alone saves about half. The saved craft passed about 10 m closer to other-level craft but made no contact in the next 10 seconds. A counterfactual on one craft at a time; nothing is switched on.
Provisional
CRAFT / 40
2026-10-12
Eleven held heavy haulers, chosen by fixed rules from the previous cycle (so not a rate), were each flown six ways under two autopilot plannings. Changing only the speed at which the simulator lets them go, from rest to cruise, changed every ending, and the hauler then ended like a normal arrival in 21 of 22 cases. Letting it go a third of a second later changed one ending; the pad clearance check and bookkeeping changed none. Nothing is switched on.
Provisional
LINK / 35
2026-10-11
At one simulated street crossing, the same four one-way lanes were re-stacked over the same four heights. The danger did not stay with one lane: on each of three stackings one lane took almost all the in-band wall losses, and it was a different lane each time. A fourth stacking lost almost no craft at all. Why is not yet tested: this is a clue for one crossing in calm air, not a recommendation, and nothing is switched on.
Provisional
CRAFT / 37
2026-10-11
With the steady cross wind, gusts and turbulence together, a hauler with a quicker assumed response made about 5 bad landings per 100 at 1.5 m/s where the assumed slow hauler made 47. Under the steady wind alone, quicker autopilot planning on the slow machine was already enough; with gusts, the planning did about 60 per cent of the work. Both response figures are assumptions, not sourced for a real craft.
Provisional
CITY / 37
2026-10-11
Turning off one controller term, the one that makes a craft match the velocity of craft on the neighbouring height level, kept saving craft wherever the term could act: with exact height knowledge, under a 4 m own-height error, on wider lane spacing and in the demonstration fleet. The price is band-keeping: craft spent somewhat more of their flying time outside their assigned height band. Nothing is switched on; whether to change the default is an open decision.
Provisional
CRAFT / 36
2026-10-11
Under a steady side wind alone, bad landings rise as a ramp from 1.35 m/s to about one in four at 1.5 m/s. Giving the same hauler a quicker response, or a gentler final slope, left none in 1,720 landings at 1.5 m/s. In this model the loss belongs to an assumed response time that has not been sourced for a real craft.
Provisional
CRAFT / 35
2026-10-10
It depends on the wind. At 1.5 m/s the steady cross wind alone, with nothing random in it, spoiled one landing in four and carried about half of the full package’s rise; gusts and turbulence added about a quarter each and no reinforcement between them was found. At 1.25 m/s the steady wind alone spoiled none.
Provisional
CITY / 35
2026-10-10
Nearly every mid-air contact under altitude assignment followed a grant made while the pair was already inside its safety distance. Withholding exactly those grants, with knowledge no craft has, removed most of the excess and kept a smaller loss reduction.
Provisional
CRAFT / 34
2026-10-05
Between the tested 1 m/s and the 2 m/s that stopped the pad, bad endings rose from about 10 to 26 to 46 per 100 scheduled landings. The station-keeping setting kept its cut at every dose.
Provisional
LINK / 28–29
2026-10-05
Not the avoidance push. Speed-matching with the other level carried the lost craft across the lane; switching it off removed the wall losses in the tested crossing, at the price of more band exits.
Provisional
CITY / 34
2026-10-05
Fewer lost craft can conceal a shift from wall strikes to mid-air contacts.
Provisional
CRAFT / 33
2026-10-04
A landing controller helped in the tested disturbance. A slower final approach made outcomes worse.
Provisional
LINK / 24–27
2026-10-04
Increasing declared height uncertainty can send avoiding craft into the buildings around them.
Recorded finding
CAMPAIGN / 268
2026-09-23
268 research cycles mapped the limits of the rule set. The next questions needed buildings.
Campaign closed
CYCLES 267–268
Historical record
Mixed: wind from 250° and 160° protected, wind from 70° did not reliably; the reason is unknown, and the research stopped here.
learned
CYCLES 265–266
Historical record
Good news: wind from 250° cut losses to a third to a half of calm at 600 and 900 craft; hints of harm failed.
learned
CYCLES 263–264
Historical record
Unclear: colliding pairs pushed apart far less than survivors, but the gap did not widen with delay, and too few collisions were seen to decide.
unclear
CYCLES 261–262
Historical record
We learned that more traffic kills through more meetings, while delay kills by making each meeting deadlier; the link study was then closed.
learned
CYCLES 260–260
Historical record
Bad news: the delay cost held at 300, 600 and 900 craft, and tripling the traffic multiplied losses about six times.
learned
CYCLES 257–259
Historical record
We learned that delay does not add close approaches; it makes each one likelier to end in a collision, up to 4.8× at 2 s.
learned
CYCLES 256–256
Historical record
Bad news: yes. In wind a 1 s delay still lost 2.7× more craft than perfect information, though wind lost fewer craft overall.
learned
CYCLES 255–255
Historical record
Bad news: losses rose steadily with every step of delay, from 0.05 per craft-hour with perfect information to 0.71 at 3 s.
learned
CYCLES 253–254
Historical record
Bad news: with a 1 s link delay ordinary traffic lost 3.8× more craft than with perfect information, and the delay is what drives it.
learned
CYCLES 251–252
Historical record
No clear difference: the emergency craft that others give way to did not spend measurably longer deep inside bubbles than a plain one.
no change
CYCLES 247–250
Historical record
Mixed: the sideways offset carries over from trip to trip, but resetting it at each trip did not clearly help safety and slowed the craft.
learned
CYCLES 244–246
Historical record
We learned the slowdown comes from matching the speed of slower traffic; switching that off doubled its speed but put it far off its line.
learned
CYCLES 238–243
Historical record
We learned that cutting the far-reaching part of the avoidance push lifts the low layer, but why it works stayed open.
learned
CYCLES 234–237
Historical record
Partly: the smaller spacing rule removed about two thirds of the sinking but brought craft much closer together; no safety conclusion.
learned
CYCLES 228–233
Historical record
We learned that a low traffic layer of wings sinks toward the ground under the layer above; the instant-response craft did not sink.
learned
CYCLES 226–227
Historical record
Mixed: a stronger pull toward the route, sized to the craft's own speed, kept it near its line, but its trips got slower.
unclear
CYCLES 223–225
Historical record
We learned that every full-speed trip flew above, below or far beside the traffic; held among the lanes, the craft slowed down.
learned
CYCLES 221–222
Historical record
No clear difference: the simple emergency craft was lost as often as a fast craft without priority, though that falls short of proving equal safety.
learned
CYCLES 215–220
Historical record
We learned that a small clockwise turn built into every dodge seems to set the direction of the drift; a candidate cause, not proven.
learned
CYCLES 212–214
Historical record
We learned that the fast emergency craft drifts north-west, more at higher speed; giving back speed-matching kept it close but made it slow again.
learned
CYCLES 208–211
Historical record
We learned that most losses on file came from a simulation artefact: other craft reappearing at their lane start just ahead of the emergency craft.
learned
CYCLES 205–207
Historical record
We learned that the fixed-wing emergency craft spends much of its flight far below its planned 60 m height, with or without priority.
learned
CYCLES 204–204
Historical record
Mixed: the rules held from 300 to 900 craft, but on a 1-second link the emergency craft kept being lost.
learned
CYCLES 202–203
Historical record
Taken back: wind protected the traffic from one direction only; from the opposite side, losses rose above still air.
withdrawn
CYCLES 199–201
Historical record
We learned that the no-radio cases had been flown without wind; flown in wind, the emergency rules cost the other traffic more.
learned
CYCLES 197–198
Historical record
Mixed: the emergency rules failed the stage's own adoption rule, but the owner made them the default anyway; they were replaced later.
unclear
CYCLES 195–196
Historical record
Mixed: the rules passed 11 of 12 test cases, but at twice traffic speed over a radio link the emergency craft was lost more often.
unclear
CYCLES 193–194
Historical record
Good news, with a caveat: over a poor radio link, the emergency craft survived as often as without priority only with full-strength dodging.
better
CYCLES 192–192
Historical record
Mixed: the emergency craft was always on time and never lost, but the rules qualified on 5 of 8 test cases, one short.
unclear
CYCLES 190–191
Historical record
Good news: with speed-matching off and a wide reserved corridor ahead, the emergency craft was on time and survived, with ordinary losses within noise.
better
CYCLES 189–189
Historical record
We learned that the delay comes from the emergency craft matching its neighbours' speed; weakening that recovered about 70 % of the lost time.
learned
CYCLES 186–188
Historical record
No difference: with right of way the emergency craft still took about twice as long as a straight flight, no faster than without it.
no change
CYCLES 183–185
Historical record
Good news: slow craft now have exact-replay checks, the loop stopped itself with nothing left, and the wing switched to full room.
better
CYCLES 180–182
Historical record
Good news: the slow-responding test craft and multirotor switched to the full extra room; the eVTOL kept three-quarters; the wing went to the researcher.
better
CYCLES 178–179
Historical record
We learned both apparent costs of the full extra room were chance: on fresh layouts each result reversed.
learned
CYCLES 172–177
Historical record
We learned all four slow airframes qualify for the full extra room; the wing's test sets clearly favoured it.
learned
CYCLES 168–171
Historical record
Mixed: the full room kept most of the crash cut and removed the storm closeness, but each airframe's case was a trade for the researcher.
learned
CYCLES 166–167
Historical record
We learned the two parts split cleanly: the smaller extra room adds closeness in storms, and early reaction removes crashes.
learned
CYCLES 163–165
Historical record
We learned the wing's speed penalty is milder in storms (2.5× against 11× in calm), and scenario types pull in opposite directions.
learned
CYCLES 160–162
Historical record
We learned the extra room's growth with speed is not the cause; more agility helps, but no tested setting flattens the growth.
learned
CYCLES 157–159
Historical record
We learned the growth rides on more close meetings at speed; a serious meeting actually ends in a collision less often.
learned
CYCLES 153–156
Historical record
We learned crossing pairs and crowded-airspace pairs close at similar speeds, so room based on closing speed had nothing to act on.
learned
CYCLES 149–152
Historical record
We learned the multirotor's storm gain is real but varies by layout, and losses still grow with speed, so the literal target was missed.
learned
CYCLES 146–148
Historical record
Good news: the gain held on fresh layouts and the setting became the default, at the price of fewer but longer close calls.
better
CYCLES 143–145
Historical record
Good news: three-quarter extra room with early reaction at strength 5 was the first setting to pass the agreed adoption rule.
better
CYCLES 137–142
Historical record
Mixed: it improved every test set but failed the agreed adoption rule, because not every crossing condition halved its losses.
learned
CYCLES 134–136
Historical record
We learned it is a trade: more extra room means fewer storm collisions but more crossing losses, and no size satisfied both.
learned
CYCLES 133–133
Historical record
We learned the extra room is what protects in storms; early reaction with the room kept improved every test flown.
learned
CYCLES 130–132
Historical record
Bad news: without the extra room, the wing's storm-test collisions more than doubled, so this version was rejected despite big crossing gains.
worse
CYCLES 127–129
Historical record
We learned early reaction works best on steering only, at strength 3; applied everywhere and stronger, it backfired at top speed.
learned
CYCLES 125–126
Historical record
Mixed: reacting earlier cut the wing's losses to about a third at higher speeds, but pairs spent longer close together.
unclear
CYCLES 121–124
Historical record
We learned lag is a large part of the wing's loss at speed, but without lag it still loses craft the ideal craft does not.
learned
CYCLES 118–120
Historical record
Bad news: every slow-responding craft loses more at higher speed; the wing loses 5× more craft at double the speed.
learned
CYCLES 113–117
Historical record
We learned the summary kept reviving withdrawn claims; after 12 corrections, both reviewers agreed the next job was flying the wing faster.
learned
CYCLES 110–112
Historical record
We learned the lane findings hold at half the traffic, where wing losses drop 8–19×; at 1,600 craft, no wind makes the wing safe.
learned
CYCLES 105–109
Historical record
No change: a faster pull back to the lane trimmed straggling but cost craft, and no setting reliably cut the wing's losses.
no change
CYCLES 102–104
Historical record
We learned stronger heading copying pushes more craft far off lane, and many straying episodes begin soon after a lane restart.
learned
CYCLES 99–101
Historical record
We learned the early neighbours are stragglers far off their own lane, and heading copying is what lets craft stray that far.
learned
CYCLES 95–98
Historical record
We learned the heading-copying rule causes the weave, centred on crossings; but for the wing that same rule also prevents losses.
learned
CYCLES 92–94
Historical record
We learned there is no jam at crossings: whole lanes crawl and weave, a behaviour of the rules themselves, worst for the wing.
learned
CYCLES 88–91
Historical record
Bad news: the wing's collision rate roughly doubles with every 300 extra craft, and no traffic level tested was safe, even with the look-ahead.
learned
CYCLES 86–87
Historical record
Good news: the look-ahead cut losses 60–81 % for eVTOL and multirotor too, and removed most wing losses at every wind tested.
better
CYCLES 83–85
Historical record
Good news: one reaction time held up on fresh layouts and became the default on 2026-09-14, though dense traffic stayed far from safe.
better
CYCLES 80–82
Historical record
Mixed: looking two reaction times ahead doubled collisions in one turbulent storm scenario, but one reaction time measurably harmed no craft type.
learned
CYCLES 77–79
Historical record
Good news: looking ahead by the craft's reaction time cut slow-craft losses sharply, though very dense traffic stayed a catastrophe.
better
CYCLES 74–76
Historical record
We learned the wing's slow response costs it most: removing the delay cut calm losses about 90 %, and wider spacing helped far less.
learned
CYCLES 72–73
Historical record
No: keeping the wing close to its lane did not reduce its losses, so the problem looked like its slow dodge at crossings.
learned
CYCLES 70–71
Historical record
We learned the wing loses five times more craft in calm than in wind, mostly between perpendicular lanes and early in the run.
learned
CYCLES 69–69
Historical record
We learned that gusts add a little close-call time for the wing, but that does not explain why it loses craft.
learned
CYCLES 65–68
Historical record
We learned that the into-wind lane drops below a tenth of calm deliveries once the wind reaches about 0.86–0.89 of cruise speed.
learned
CYCLES 61–64
Historical record
Unclear: the wing's extra offset is a real resting point, but neither simple explanation fits and the cause was not found.
unclear
CYCLES 59–60
Historical record
Good news: the crosswind arithmetic held within 2 m when the no-pull zone changed; the proposed reason for the wing's extra 45 m failed.
learned
CYCLES 55–58
Historical record
We learned that ideal craft settle where simple arithmetic predicts; the wing settles about 45 m farther out, and strong crosswinds still win.
learned
CYCLES 51–54
Historical record
Bad news: wind blows whole lanes sideways, by kilometres in a strong crosswind, because craft do not correct for drift.
learned
CYCLES 48–50
Historical record
Bad news: there is no free setting; each weaker speed-matching setting keeps lanes better but loses more crowded wings.
learned
CYCLES 45–47
Historical record
Mixed: speed-matching is what pushes wings off their lanes, but switching it off costs many crowded wings.
learned
CYCLES 43–44
Historical record
Good news: the deeper close calls were all test artefacts from putting craft back at lane starts; with that fixed, none remain.
learned
CYCLES 41–42
Historical record
No difference: on wings that cannot stall, the anti-braking switches showed no clear saving in craft and pushed the fleet farther off its lanes.
no change
CYCLES 38–40
Historical record
Good news: the very deep close calls came from the test itself, which put craft back at a lane start on top of others.
learned
CYCLES 36–37
Historical record
We learned that the 3 m/s wing needs both anti-braking switches; whether they make wings overshoot crossings stayed unanswered.
learned
CYCLES 33–35
Historical record
Mixed: stopping two rules from braking nearly ends the wing's slow time and saves stalling craft, but survivors spend more time in close calls.
unclear
CYCLES 31–32
Historical record
Mixed: stopping the speed-matching rule from braking the wing removed about half of its slow time, and saved some stalling craft.
unclear
CYCLES 28–30
Historical record
Mixed: the speed-matching rule is what slows the wing, but in crowds it prevents collisions, so it stays on.
learned
CYCLES 25–27
Historical record
We learned what it is not: slow moments rarely have traffic nearby and the controller never asks to slow down; the cause stayed unknown.
learned
CYCLES 24–24
Historical record
Bad news: a fixed wing spends a quarter of its time at minimum speed, and if that meant stalling, most of the fleet would crash.
learned
CYCLES 22–23
Historical record
We learned that collisions follow the sideways pull back to the lane, not the vertical one, though how it causes them is unproven.
learned
CYCLES 19–21
Historical record
Mixed: weakening the pull back to the lane cut collisions steadily, but the crowded sky stayed unsafe at every strength.
unclear
CYCLES 18–18
Historical record
We learned that the ±12 m band is simply where the pull back toward the lane holds craft, not room needed for dodging.
learned
CYCLES 17–17
Historical record
Bad news: neither way of sizing the up-or-down dodge to the airframe kept lanes at their height; both were rejected.
worse
CYCLES 15–16
Historical record
Bad news: dropping the vertical dodge made things worse still, and only a very tall lane wall was as safe as no wall.
worse
CYCLES 12–14
Historical record
Bad news: forbidding up-or-down dodges inside a lane made every test sky far less safe, the worst change tried so far.
worse
EARLY / F
Historical record
Bad news, then fixed: four measurement faults were corrected, and a stricter rule for calling anything safe was written.
learned
EARLY / F
Historical record
We learned most collisions are crossings away from the pads, and one lane crossing accounts for a third of them.
learned
EARLY / F
Historical record
Taken back: with six layouts, half of the interesting claims changed, including the storm-plus-slow-link interaction.
withdrawn
EARLY / E
Historical record
We learned, on this test set, that time-related limits moved the safety line while force-related ones cost deliveries instead.
learned
EARLY / E
Historical record
Mixed, and partly taken back later: on two layouts, crowding and link quality multiplied realistic craft's losses, while weather did not.
learned
EARLY / E
Historical record
We learned wind stops deliveries at a set ratio to cruise speed, without collisions, and spacing settings matter only in a narrow range.
learned
EARLY / E
Historical record
Good news: for ideal craft, no single weather, crowding or speed setting caused a collision; only the link and agility limits did.
learned
EARLY / D-cont
Historical record
We learned the 'fixed wing' was not a real wing, and one vertical spacing question had never been tested.
learned
EARLY / D-cont
Historical record
Good news with a steep price: spacing sized to each craft's weakness won back most extra collisions, but cost a fifth or more of throughput.
better
EARLY / D-cont
Historical record
Bad news: realistic agility limits raised collisions sharply; response lag hurt most, because the right dodge arrives too late.
learned
EARLY / D-cont
Historical record
Bad news, confirmed: an outside review found real defects in the wind model and the scoring; testing paused to fix them.
learned
EARLY / D-cont
Historical record
Unclear: a direction-aware wind launch rule added trips without collisions, but was never finished or adopted.
unclear
EARLY / D-cont
Historical record
Bad news, then fixed: three pad scenarios had counted zero trips, hiding any loss of deliveries.
learned
EARLY / D-cont
Historical record
Bad news: steering against measured gusts cost collisions; steering on the forecast wind helped tracking but was never finished.
worse
EARLY / D-cont
Historical record
No gain: neither fix for craft pushed to the ground in storms helped; one did nothing and one made things worse.
unclear
EARLY / D-cont
Historical record
Good news: tying each craft's height to its heading cleared crossing-angle impacts and roughly halved one sweep's losses.
better
EARLY / D-cont
Historical record
Good news: slowing fast craft that close in faster than the link can warn cut fast impacts from 61 to 4.
better
EARLY / D-cont
Historical record
Good news: skipping pairs that cannot matter made storm-on-slow-link runs up to 2.5 times faster, with identical results.
no change
EARLY / D-cont
Historical record
Good news: making the required spacing grow with gust drift over the link delay cut spacing-test impacts from 1,796 to 66.
better
EARLY / D-cont
Historical record
Good news: giving neighbours shared gusts, as real air does, removed more than half the storm-on-slow-link losses.
better
EARLY / D-cont
Historical record
We concluded at the time that storm-on-slow-link losses were set by the wind model, not the rules; a wider margin made them worse.
worse
EARLY / D-cont
Historical record
Bad news: the learned margin for old news raised collisions from 44 to 75, so the hand formula stays.
worse
EARLY / D-cont
Historical record
Good news: a small learned model predicted well which untested settings break, and its picks found hundreds of impacts.
learned
EARLY / D
Historical record
Good news: two fixes to how fog shortens ranges removed most fog-and-storm collisions.
better
EARLY / D
Historical record
Good news: letting crowded craft give up their goal in extreme storms roughly halved storm collisions without losing work.
better
EARLY / D
Historical record
Bad news: sending intentions or acceleration so others could predict ahead made things worse; the simple straight-line guess stays.
worse
EARLY / D
Historical record
Good news: craft near a neighbour sending four times as often cut slow-link collisions from 13 to 5.
better
EARLY / D
Historical record
Bad news, then partly fixed: craft strayed far outside their lanes; a pull back to the lane and a dodge-direction rule helped.
better
EARLY / D
Historical record
Bad news, then fixed: a recent rule had raised slow-link collisions from 37 to 311; two fixes brought them down to 16.
better
EARLY / D
Historical record
Good news: fixing how queued craft swap places removed the last still-air collision; the full test set reached 0.
better
EARLY / D
Historical record
Good news: six pad changes took the wind tests from 15 collisions to 0 and let crosswind pads land craft again.
better
EARLY / D
Historical record
Good news: giving avoidance priority on predicted near-misses cut the full test set from 7 collisions to 2.
better
EARLY / D
Historical record
Good news: extra spacing for old news, used only between fast-moving pairs, more than halved collisions on the slow-link tests.
better
EARLY / D
Historical record
We learned where the link breaks: fine at 2 messages a second, failing at 1 a second with half lost and 2 s delay.
learned
EARLY / D
Historical record
Good news: allowing for how far gusts could push each craft cut turbulent stream collisions from 22 to 1.
better
EARLY / D
Historical record
Good news: letting each queued craft move only after the one ahead has left cut pad collisions, for about 5 s of waiting per move.
better
EARLY / D
Historical record
Good news: a string of pad faults was found and fixed, and the full test set fell from 329 collisions to 33.
better
EARLY / D
Historical record
Good news: a package of fixes cut one hard test group from 198 collisions to 2; the landing pads then held nearly all losses.
better
EARLY / C
Historical record
Good news: releasing craft to the pad one per second cut close calls by about two thirds, at the cost of slower landings.
better
EARLY / C
Historical record
We learned the danger is not a growing wave: disturbances still faded, but close calls multiplied where tight spacing met delay.
learned
EARLY / C
Historical record
We learned that at high crowding safety runs out before capacity: traffic kept flowing, but craft came too close.
learned
EARLY / B
Historical record
Bad news: mixing small, cargo and heavy craft at one height led to many deep close calls.
learned
EARLY / B
Historical record
Good news: a craft that went silent was avoided with an even wider berth than one that kept broadcasting.
learned
EARLY / B
Historical record
Bad news: the crowd converging on the landing pad is where safety runs out; the flight in was nearly clean.
learned
EARLY / B
Historical record
Good news: no jam up to 300 craft per km²; speed halved, but the total traffic moved kept rising.
learned
EARLY / B
Historical record
Mostly good news: radio delay weakened the damping but never made the disturbance grow in this test.
learned
EARLY / B
Historical record
Good news: as predicted, stronger speed-matching damped the shove more, and the disturbance always faded.
learned
EARLY / B
Historical record
Good news: a shove to one craft faded as it passed through the grid, in every version of the rules tested.
learned
No flights match those filters. Try a broader question.
Earlier entries describe the model at the time of each experiment. Later amendments and corrections are retained alongside the findings.