Thirty-three and forty-two. Those two numbers are the whole story. A preprint posted to arXiv on 28 August 2026 — The scales of urban mobility shape socioeconomic mixing, by Andrew Renninger, Egor Kotov and Carmen Cabrera — decomposed roughly 200 million trips from about 20 million GPS devices across 226 functional urban areas in France, Germany and the UK. Its finding, verbatim: distal trips, the longest class, account for 33% of travel but facilitate 42% of inter-group encounters. Nine percentage points of over-performance. It is a preprint, not yet peer reviewed — and the first time anyone has priced what proximity planning quietly spends.

Now let the slogan walk in. Every compact-city instrument written in Europe since 2020 — proximity planning, polycentric development, the 15-minute city, Zurich’s own short-distance logic — scores its success in kilometres saved. This paper puts a second currency on the table: social exposure, read as a Hill number of order 1, the effective number of income groups a resident actually encounters. The two currencies are in tension.

The authors fit a Gamma mixture to each neighbourhood’s trip-length distribution and let three scales fall out — proximal, medial, distal. The proximal scale, the radius of genuinely local life, is nearly identical in every neighbourhood of every city measured. The 15-minute neighbourhood is not a thing to be built; on this evidence it already exists, without policy, at constant radius in Marseille and Munich. What differs between an inner and an outer address is not the size of the local world but how far the trips that leave it must travel.

Then the control any honest write-up owes the reader. Hold every neighbourhood’s trips and their lengths fixed, but send them to random reachable destinations, and the city mixes 19–47% more than it actually does — 19% in London, 24% in Paris. Distance is not the binding constraint. Choice is. The localisation debate — should we shorten trips? — argues about the smaller lever. Reallocating distal trips down to the proximal scale costs 25–36% of experienced mixing (25% in London, 36% in Paris); the “up to one third” of the headline is the top of that range, not the average. The milder, medial reassignment costs about 11%. Either way, you spend mixing to buy kilometres.

←TODAY: the preprint is open-access CC BY and the Zurich income and flow datasets are free downloads — the whole computation is a Monday-morning job. →3012: the city that lasts shortened the journey without shrinking the reach. Fulcrum: distance is a cost and a product at once; the trade only shows when you price both directions.

Be honest about the gap: no Swiss city is in the dataset — France, Germany and the UK only. Anyone writing “Zurich’s distal trips carry 42%” has invented it. What Zurich has is the apparatus to run the test on itself. The ETH hook is fresh: on 31 July 2026 the Institute for Transport Planning and Systems (IVT) published its Sustainable Urban Transitions Lab work, led by Mariana Montero Vega, on first-mile/last-mile micromobility in Hannover and Seville. Read alongside Renninger’s result, first-mile/last-mile stops being a convenience layer — it is the medial mechanism that keeps a distal trip alive without a car. The criticism goes at the instrument, not the planners: Swiss compactness is scored in Verkehrsleistung and land consumption, with no line item for social exposure. PAZ has covered the systemic version before — our Embarcadero Freeway read, where infrastructure optimised for one variable failed on every variable left outside the model.

Atelier: On Monday this changes one habit — the ground-floor programme question. An office running a Quartier study or a mixed-use-plinth competition currently justifies that plinth on walkability. The paper hands you the opposite justification for the other move: the destination that pulls people in from three Kreise away is not a failure of compactness — on this evidence it is where the mixing gets made. The Monday move: bring both arguments to the next Vorprüfung as an explicit trade, and let the client choose a regional-draw programme knowingly.

Hack: Count the effective number of income groups a site can reach at each radius; watch it climb as the ring widens — that climb is what localisation spends. Load Statistik Stadt Zürich’s geo_statistische_quartiere and fd_median_einkommen_quartier_od1003 into PostGIS, rank the 34 quarters into five quintiles with ntile(5) over (order by median_income), then buffer your site at 1 km, 4 km and 15 km. For each ring, take the exponential of the Shannon entropy of the area-weighted quintile mix — the paper’s Hill number of order 1:

SELECT ring, exp(-sum(w * ln(w))) AS effective_groups
FROM (SELECT ring, quintile,
             area / sum(area) OVER (PARTITION BY ring) AS w
      FROM rings_x_quartiere) t
GROUP BY ring ORDER BY ring;

The 15 km ring will report more effective groups than the 1 km ring. That delta is the social range a fully localised city trades away — now visible, not assumed.

Run it. The source is four days old, its central claim is a comparison of two percentages, and the Zurich data to test it is free to download, already cut into 216 zones. Draw the real reach graph of your site at three radii before you sign off on compactness as an unqualified good — the third scale you were about to shorten may be the one doing the social work.