London feeds one runway from four holding stacks ringed around the city. Whatever corner an arrival comes down through (holding when the flow is heavy, routed straight in when it is not), it merges onto the same single final, and the airplanes land about ninety seconds apart, over and over. This is an attempt to reverse-engineer the rules the Director builds that gap with, read off public aircraft motion: what sets it, what does not, and where on the approach it is made.
Reconstruct every landing pair over four westerly days from public aircraft motion, measure the gap at a gate on short final, and the result is a single tight lump. The median gap is about ninety seconds. About three in five land between 75 and 105 seconds. There is one mode, not several. Ninety seconds is the number the Director works to. The rest of the page investigates how.
A single mode is already a clue. If the gap were an accident, a leftover of whatever timing or traffic happened to arrive, it would smear. It does not. Something is building it to one value. The sections that follow take that apart.
The obvious guess is release timing: space the aircraft out of the stacks and that spacing carries to the runway. Watch when each aircraft is released from its stack and the intervals are all over the place, back to back one minute, three minutes apart the next, sometimes even out of order (the pair swaps on final, so the gap goes negative). Watch when those same aircraft land and the gaps collapse to ninety seconds.
Even the stack an aircraft comes from, which fixes its rough distance to the runway, does not set the gap. BIG is about four and a half minutes out, BNN about nine. That gradient decides who lands first; how far apart is set later, on final.
The Director does apply wake separation, so rather than assume the rule, we read it off the deliveries. Tag every landing with its aircraft type, group the consecutive pairs by leader-and-follower wake category, and take the median gap each pairing actually lands with. If a pair-by-pair distance table drove the gap, the medians would fan into well-separated bands, one per pairing.
Two rules fall out. First, wake does step the delivered gap, but gently, and the expensive pairings land tighter than the generic table would ever allow. Second, from the landing sequence itself: the expensive heavy-ahead-of-medium pairing is under-represented, about 16% of pairs against 22% under independent ordering. Part of that is wake-aware sequencing set upstream, part is just Heathrow's banked schedule; either way the worst pairings are thinned before they reach the runway. So the tight single mode is not wake-blind: a small separation premium flown tight, with the expensive pairings kept scarce in the order.
To find where the number is made, measure the same pairs not just at the gate but at every distance to touchdown. Sample the gap between two aircraft while they are still forty, thirty, twenty miles out, and watch when it settles. Leaving the stacks it is chaotic. It collapses over a few miles of final and then holds, unchanged, to the runway.
The approach control team works it with two levers, in series: first the track, a tight dog-leg onto the ILS that sets the gross gap, then the speed, which does the fine spacing once an aircraft is established on the localizer.
Section 4 named two levers, the track and the speed, and they act in different places. The gap is essentially made before the final turn: once an aircraft is on the localizer it changes by only a couple of seconds, though a median dog-leg of about 1.3 times the straight-line distance shows the track was working hard to get there. What the speed does shows up on the localizer itself. The exact share each lever takes is entangled with where each aircraft was released, so the motion cannot split it cleanly, but where the gap settles, and what the speed does there, are clear.
Read off the motion, the whole operation comes down to a few habits. The Director works to one gap of about ninety seconds and rebuilds it on final, whichever stack an aircraft came down from. Wake still steps the gap, but gently, tighter than the book, and the expensive pairings are thinned from the sequence before they reach the runway. None of it is a clock counted at the stack or a table read at the gate: it is one gap, built over and over by vector and speed on final.
MethodFour westerly days (26 to 29 June 2026), a wide capture around the field. One landing runway (London runs segregated, so every arrival is a single in-trail stream). 2382 landing pairs, the 2019 with a measured release plotted in the scatter and its fit. A release is an aircraft leaving its stack; the delivered gap is measured at a fixed gate on final, and for the convergence figure the same gap is measured at successive distances to touchdown.
The wake rule is measured, not assumed: each landing carries its ICAO type, so the delivered median is grouped by leader-and-follower wake category; the "book" it is shown against is the generic ICAO Doc 4444 4-category distance minima (radar minimum where no wake pair applies) divided by the follower's own measured final groundspeed.
Our caveats: (1) that generic book is not the finer, tighter RECAT-EU / time-based scheme Heathrow actually flies, so it overstates the real minima, which is exactly why the delivered gaps beat it on the expensive pairs; (2) the gate used here is short final, not the threshold, a proxy taken because speed is roughly stabilised there; (3) the stream is largely wake-homogeneous (most pairs are same-category), so part of the single mode is the traffic mix, not the controller alone. The right tail (about 6% of pairs land more than 150 s apart) is mostly ordinary same-category pairs in a demand lull, plus a handful of go-arounds, not a wake effect.
Anyone with an equivalent ADS-B pull can reproduce or refute every figure here. These are correlations, a description of the delivered stream, not a controller's rulebook or an operational display.
DataAircraft positions from adsb.lol globe_history, licensed ODbL 1.0.