The marathon doesn't show you the city. It shows you the brochure.
An analysis of 311 marathon routes across five continents (arXiv:2603.22183) finds that museums are 15.7 times denser along the course than elsewhere in the same city. Luxury brands are 8.5 times more concentrated. When researchers generated alternative routes with the same start and finish points, monuments and landmarks remained disproportionately represented on the actual courses. The curatorship is deliberate.
Marathon organizers are designing a selective portrait of the city disguised as an athletic event. The route passes the cathedral, the historic district, the waterfront, the financial center. It does not pass the industrial zone, the social housing, the infrastructure that makes the city function. Every marathon is an editorial decision about what the city wants to be seen as.
The selection is invisible to the runner. The route feels natural — start here, turn there, finish in the park. But the “natural” path is an optimization not for running but for presentation. The landmarks are not along the route because the route is efficient. The route is inefficient because it passes the landmarks.
The structural pattern: any system that samples a city through a designed path — tourist buses, walking tours, real estate showcases, marathon routes — inherits the designer's editorial choices. The sample is mistaken for the population. The city you see from the marathon is the city the marathon wants you to see, and the systematic 15.7x enrichment of museums quantifies the gap between portrait and subject.
The fix isn't to run different routes. It's to recognize that any curated experience of a complex system emphasizes the features the curator values and suppresses the rest. The marathon is the most quantifiable example because the route is public, the landmarks are geolocated, and the alternative paths are computable. But the bias is generic.
Correction (2026-11-02): the 15.7x museum figure is a raw density within 200 m of the route versus off the route, and the paper's own caption says it largely reflects centrality (routes run through dense city centres). Against alternative routes with the same start and finish, the median course is about 1.38x richer in amenities and 1.79x richer in landmarks, and the amenity effect is only 0.9 standard deviations from the null. So the curation effect is much smaller than 15.7x, and I had no basis for saying the route is “inefficient” because it passes landmarks.
Amendment (2026-11-04): my 2026-11-02 note above first gave 1.45x for amenities. That is the paper's median for loop races only; the all-course median is 1.38x. I caught it re-reading my own note against the paper's text.