The question
Earlier entries found that an honest height allowance makes the crossing gate refuse many pairs, after which hundreds of craft were lost to walls. What actually moves a refused craft into a wall in its last seconds?
The flight
Cycle 28 recorded, for every lost craft, what pushed it in its final seconds and what it hit; its registered loss floor failed on one set, so that cycle is reported as a description only. Cycle 29 added a telemetry-only recording switch, off by default, decomposed the controller command over each lost craft’s last 10 s into its terms on 24 fresh seeds per set, and then switched single terms off globally as knockouts.
What happened
Every loss was a flight into a building side. Over the last 10 s, alignment, the term that matches speed with craft on the other lane or level, pushed toward the wall by 1.85 to 2.15 times the craft’s actual across-lane displacement on all four live sets, partly cancelled by the lane-heading and pull-back terms; the pair push from refused partners contributed at most 0.11 of that displacement. Switching off across-lane or cross-band alignment removed every wall loss at the 4 m allowance on the standard link (323 to 0) and cut them to 10 to 26 per cent at 8 m, with no mid-air contacts added. The vertical dodge is the escape, not the cause: removing it roughly doubled the losses.
What we learned later
Amendment (2026-10-10, LINK cycle 30, provisional). The alignment result replicated on 48 fresh layouts: with speed-matching between height levels off, the standard tier lost 150 craft to walls at the 8 m allowance against 728 with it on. The leftover losses were then traced to the refused pair’s own dodge. An off-by-default hold at the gate lets a refused craft that is inside its height level drop the across-lane part of that dodge: wall losses fell from 150 to 63 on the standard tier and from 72 to 6 on the better tier, with no craft-to-craft collisions, for 3–5 per cent more band exits and no fewer lane completions. Clipping the vertical escape instead multiplied losses 9 to 19 times on both tiers, so the vertical dodge is how refused pairs escape, not what puts them on the wall. Without the alignment switch the hold changed nothing, and under a 4 m own-height error the loss count was identical with it on or off. Eight registered predictions hit. Nothing was adopted: the hold needs the alignment switch, which belongs to the CITY corridor rules; both stay off. Its map sentence awaits the cycle-31 review.
research/city/design/B-LINK-GATE-BACKSTOP-record.md
Follow-up (2026-10-11, LINK cycle 31, provisional). The cycle-31 review ruled the cycle-30 step flawed in its findings; they were fixed in a later section of its record. The hold was then re-flown on 48 new layouts in calm air with speed-matching between levels off: wall losses fell from 148 to 74 on the standard tier and from 76 to 3 on the better tier, with no craft-to-craft collisions, so the calm result replicates. The price is about 3 to 4 per cent more band exits; whether the hold itself raises the exit rate could not be settled, because the rise did not hold in each half of the layouts. The losses left on the standard tier are craft refused while already outside their height band, which the hold by design does not touch. The wind question has no answer yet: under the project’s standard cross wind every one of the 1,728 craft in each run was lost to buildings with the hold on or off, so the hold cannot be judged in wind until craft can keep station between buildings. Five of eight registered predictions hit. Nothing was adopted; the hold stays off.
research/city/design/M-LINK-HOLD-WHY-record.md
Follow-up (2026-10-12, LINK cycle 32, exploratory: the measurement plan was not reviewed before it flew). The clue above, that the losses the hold leaves are craft already outside their height band, was checked against time. Craft spend about 80 per cent of their flying time outside their 6 m band, so many refusals happen there anyway. Per second of flying, without the hold, a craft inside its band was about 5 times (standard tier, 5 updates a second) to 20 to 100 times (better tier) more likely to hit a building than one outside it, on 48 new layouts and on cycle 31’s. The hold works because it acts on craft still in their band: on the standard tier it cut in-band losses from 80 to 13 and out-of-band losses from 77 to 48. No larger outward push was found before a lost craft’s last departure from its band, and the standard tier pushes craft across bands about 17 per cent harder per second outside the band than the better tier. One of four registered predictions hit. Nothing was adopted; the hold stays off; the sentence for the research map is proposed for the next review.
research/city/design/M-LINK-OUTSIDE-WHY-record.md
Follow-up (2026-10-12, LINK cycle 33, provisional). The cycle-33 reviewers found the cycle-32 wording overclaimed and it was corrected in its record: its in-band versus out-of-band rates were crude, not adjusted for where craft fly; its tier push compared what the controller commands, not the craft’s actual motion; and it withdraws the cycle-31 clue without explaining it. Cycle 33 then compared craft at the same place (same lane, same distance to the nearest crossing, same side), on two new sets of 48 layouts, without the hold, in calm air. About half of the cycle-32 gap was place: in-band craft fly more near the crossings. The rest stays: matched on place, the out-of-band over in-band loss rate per second of flying was 0.56 and 0.57 on the standard tier (5 updates a second) and 0.21 and 0.12 on the better tier, every range below 1. The risk is highest in the upper half of the band and drops just above it. In the lost craft’s last 10 seconds the sideways move toward the wall was carried by their reaction to craft in other height bands, about 3.7 times the net move, with the heading hold pulling most of it back, inside and outside the band alike. Whether a recent refusal to change height marks the lost craft could not be tested: near crossings almost every craft had one. Giving each craft the true sideways position of its other-band partners did not cut the in-band losses: no effect was detected on the better tier, and on the standard tier the two layout sets disagreed (one showed more losses). No craft-to-craft collisions. Two of four registered predictions hit; the main statistics were changed after the predictions and before any run, as disclosed in the record. These are associations, not a cause. Nothing was adopted; the sentence for the research map awaits the next review.
research/city/design/M-LINK-INSIDE-WHY-record.md
Correction (2026-10-13, LINK cycle 34, provisional). The previous note said the risk is highest in the upper half of the band. That reading did not survive a like-with-like comparison on fresh layouts, and it is withdrawn as a height effect. Almost every in-band wall loss is on one band, 45-W at 50 m (96 to 100 per cent in each of four runs); the other middle band had 0 to 7 and the outer bands none. Within 45-W the lower and upper halves lose craft at about the same rate once the lane is taken into account, on both assumed radio tiers (L3, 5 Hz; L4, the better radio) and both seed sets. The earlier peak came from that band spending about five times more of its in-band time in its upper half; the band below it spends its time in the lower half and loses nothing. The sideways push from craft in other bands comes about 60 per cent from crossing-street craft above and 40 per cent from same-street craft on the level below: a lead, not a clear winner. Whether the half facing the crossing street is worse could not be measured, because the other middle band has too few losses; a count before the run should have caught that. One of four registered predictions hit, and that one rests on an empty cell. No simulator change; nothing was adopted; the sentence for the research map awaits the next review.
research/city/design/M-LINK-UPPER-HALF-record.md
Correction and follow-up (2026-10-13, LINK cycle 35, provisional). The cycle-35 reviewers narrowed the previous note: “about the same rate” means no difference between the lower and upper halves was detected, which does not show they are equal; on the worst radio a modest upper-half excess on that band is not ruled out. The 60/40 push was measured on losses in the band’s upper half, where crossing-street and same-street craft are also above and below, so the two splits cannot be told apart there. The predictions were 0 of 3 that could be scored, with the fourth void, not one of four. Cycle 35 then re-stacked the same four lanes over the same heights: the in-band losses moved to a different lane on each of three stackings and almost vanished on a fourth, so the risk on westbound 45th is a property of the plan’s stacking, not of that lane (see “Move the lanes, move the danger”). No simulator change; nothing adopted.
research/city/design/M-LINK-WHY-45W-record.md
What this does not establish
Descriptive carrier shares on one crossing geometry and plan. The knockouts are global switches, not a corridor rule, and nothing was adopted. The record’s first reading that throughput was equal or higher was withdrawn at integration: its price field counts band exits, the times a craft’s true height leaves its band, and those rose by about half under the knockouts at 4 m. Throughput through the crossing is unmeasured. The proposed map sentence awaits the cycle-30 LINK review.
The next question
Alignment across levels at a crossing belongs to the corridor rules, so the question was handed to the CITY track. The LINK track continues with a gate backstop for the losses that remain at the 8 m allowance.
Source references for reproduction
Optional repository references, pinned to this snapshot. Repository access may be required.