The walkable model

WASD or the arrow keys to move, Q and E to turn, drag to look. Eye height is 1.62 m.

The figures walk the real network and nothing else about them is real. They are here to show that the network is connected and where it goes. Their number is a constant in the source. No count of pedestrians in this corridor has ever been made.

What this page was built to settle

Until this page existed, every position in this repository was typed to four decimal places off a basemap. Four decimal places of latitude is about 11 m. The canyon under the bridge is 91 m wide, so an 11 m error is an eighth of the width of the thing being described — and it was never measured, because there was nothing to measure it against.

There is now. Each inherited position can be tested against surveyed ground with three questions that have definite answers:

1. Does it lie inside a building? For a place the model itself marks as outdoors, that is not an approximation. It is a position that cannot be right.
2. How far is the nearest wall?
3. How far is the nearest walkable path? A place nobody can walk to is a place no pedestrian model should be spawning anybody at.

The result

2 of 18 places in the agent model are marked outdoors and fall inside a surveyed building footprint: York St (F), Washington St view. A further 2 fall inside a footprint and are marked indoors, which is correct and is why the indoor flag is carried here rather than judged.

2 of the four points the MTA instrumented also resolve inside a building footprint: Adams Street Library, Front and Pine Street. That is not necessarily an error — the memo describes indoor sessions as well as outdoor ones, and section 1.7 of the concept document already found that the indoor rows behave differently from the outdoor ones. It is recorded here because it is checkable and because nothing else in this repository could have checked it.

Nothing has been moved. Not one coordinate in the agent model was corrected by this page. A measured offset published beside the original is evidence; a quietly edited coordinate is a different guess with the audit trail deleted. The corrections belong in the model, made deliberately, with the reason recorded — not as a side effect of building a drawing.

The four measured points, on real ground

Levels are quoted from the MTA's own memo and are rated 5/5. The positions are ours, digitised from the memo's written descriptions, and are rated 2/5. This table is the first time the two ratings have been visible in the same row.

The level wash in plan view is inverse-distance weighting in the energy domain between those four numbers. It is an interpolation, not a propagation model, and the agent model records why one cannot be fitted here: jittering the digitised positions by the width of their own uncertainty moves the fitted decay exponent anywhere from 0.7 to 22.3.

The walk, routed rather than drawn

1528 m from the York Street F platform to Pier 1, computed over the real footway network by shortest path rather than drawn as a line between waypoints. It now continues along the water past Jane's Carousel to Fulton Ferry Landing before turning down to Pier 1, which is the walk people actually take and which the earlier drawing stopped short of.

1528 m and 1,482.44 m are two different walks, not two estimates of one walk. The carousel page publishes 1,482.44 m over five waypoints. This page adds a sixth, Fulton Ferry Landing, because a route that stops at the water's edge leaves out the half of the walk that goes back under the bridge — which is the half the canyon argument is about. Nothing has been corrected. The carousel figure is still right for the route it describes, and the two must never be quoted against each other as though one superseded the other.

Two references were checked first, and neither one won

This page was asked for with two examples attached: GeoLibre, an open-source geospatial platform, and Race Condition, an earlier project of ours that renders a marathon through a city in 3D. Both were read properly — source, data files, build steps — before a line of this was written. The finding has the same shape this repository keeps producing: the thing you reach for is downstream of, and worse than, the thing already sitting in the repository.

Race Condition's own backlog says it plainly: “Building heights: synthetic”. Its photoreal look does not come from its data at all — it comes from a proprietary photogrammetry export that is not in the repository. GeoLibre is a serious piece of work and the disqualifier is narrow and absolute: its Python export writes an <iframe> pointing at a hosted application, so the artifact stops working the moment the network does. Everything in this repository has to open from a file on disk with nothing fetched.

What was taken from each. From Race Condition, the pattern — an instanced city, agents constrained to a path, an anchor-relative local grid in metres, one animation loop. That pattern is sound and it is what this page is built on. From GeoLibre, one pointer: its Overture Maps layer carries building footprints with a height field, which is the plausible route to heights for the far bank, since the city file that supplies every height here stops at the New York City line.

The comparison is not a criticism of either project. Both are built for a general case. This one had a city file with surveyed roof heights already in it, and the general case cannot beat that. The transferable part is that the check took an afternoon and would have cost weeks had the answer been assumed the other way.

Where this is likely to be wrong

1. The bridge deck elevation is invented. It is drawn at 27 m because something had to be drawn. No public source gives the deck elevation over DUMBO, which is why it is the only object in the massing frame not surveyed to the roof, and why it is hatched here. Every sightline this page appears to show under the bridge inherits that number.
2. Carriageway and path widths are a drawing convention. OpenStreetMap gives a centreline, not a kerb. The ribbons are drawn at plausible widths so the plan reads. They are not a survey and no measurement should be taken off them.
3. The street network is crowd-mapped, not surveyed. The building footprints are the city's own file; the paths are OpenStreetMap. Those are not the same standard of evidence and they are drawn on the same page, which invites reading them as if they were.
4. Ground is flat at zero. The footprint file carries a ground elevation per building and it is used for the base of each wall, but there is no terrain surface between them. DUMBO rises away from the water; this model does not.
5. The figures are not a population. Their count is a constant, their speed is one number, and they choose turnings at random. The population model is a different artifact with a different and much longer list of what is invented in it.
6. Only the largest connected component is walkable here. Orphan fragments in the network were dropped so that no figure spawns on a disconnected stub. That is right for the drawing and it means the node count on this page is smaller than the raw network.
7. Across the river is decoration. Footprints there are simplified to whole metres, anything under about four storeys is dropped entirely, and none of it is walked, measured or reasoned over.
8. An offset is not a correction. This page measures how far each inherited position sits from surveyed ground. It does not establish where the position should have been, and for the MTA points it cannot, because the memo describes them in words rather than coordinates.
9. The span stops where the data stops. The deck is drawn only where bridge-carried track geometry was fetched, so it ends in mid-air over Manhattan rather than at an abutment. Nothing has been extrapolated to make it look finished.
10. The evaluation of the two reference projects is a snapshot. Both are actively developed. The finding is about the state of each on the day it was read, and the files that carry it are named so the claim can be re-checked rather than taken on trust.

Sources, and how to rebuild every line of this

• Building footprints and roof heights — NYC OpenData 5zhs-2jue, the city's building file. 1,034 in the walkable extent, 1,068 across the river. 5/5
• Streets, footways, steps, piers, parks, shoreline — OpenStreetMap via Overpass, © OpenStreetMap contributors, ODbL 1.0. 756 carriageways and 2,230 footways and steps. 4/5
• Walkable network — derived from the above: 8,712 connected nodes, 10,200 edges.
• Sound levels — MTA's own instrumented sessions, as quoted in the concept document. 5/5

Nothing here is traced off a proprietary basemap. Fetch the data with data-collection/fetch_geodata.py and redraw this page with build_walkable_map.py. Built 2026-08-11.