The Section Problem

How far down can we actually model?

The brief asks for a model that reaches "down to the schematics of screws, track rails, attachments, noise barriers, clearance." Here is the honest ladder — and where it stops.

The occlusion problem

Every citywide 3D dataset New York publishes is captured from an aircraft. An aircraft sees the top of the deck. The geometry that governs this entire research programme — the underside, the floor beams, the track stringers, the fastenings, the truss interior, the clearance envelope over the streets of DUMBO — is in the deck's acoustic and optical shadow simultaneously.

This is not a resolution limit that better LiDAR would fix. It is a line-of-sight limit. There is a hard floor at L2 for every free public dataset, and it is geometric, not financial.

Geometry sources actually checked

Each row was queried directly, not assumed to exist. Negative results are recorded because they are the expensive thing to rediscover.

The HAER result, in detail

The Historic American Engineering Record is the natural first stop for measured drawings of landmark American structures. Querying the Library of Congress collection API directly returns, for the three East River suspension bridges:

The Brooklyn Bridge has a sheet titled "Plan, elevation, detail at Manhattan tower." Its two neighbours — the only two that carry subway trains — have large-format photographs by Jet Lowe from 1991 and no drawings at all. The photographs are not worthless: they are large-format, archivally scanned, and a plausible photogrammetric input, though no camera calibration data accompanies them and self-calibrating structure-from-motion would have to recover it. But they are pictures of a bridge, not a description of one.

An all-open-source path from public data to reviewable model

No proprietary licence anywhere in the chain, so the result can be published under the same CC BY 4.0 terms as the written research and rebuilt by anyone who disputes it.

Where the toolchain will mislead you

Stated up front, because the failure mode of a good-looking pipeline is unearned confidence.

• NoiseModelling implements CNOSSOS-EU, and CNOSSOS-EU's rail source model is calibrated on European at-grade and ballasted track. A 1903 open-deck steel suspension bridge with jointed rail on stringers is outside its domain. It will still return a number.
• CNOSSOS handles a barrier by diffraction over an obstacle. It does not model a steel structure re-radiating as a surface — which the companion research argues is the central mechanism. The one thing you most want modelled is the one thing it does not do.
• Use it for relative geometry, not absolute level. "Does moving this barrier 3 m change the shadow at the dog run" is a fair question to ask it. "How many decibels" is not.
• Photogrammetry produces confident geometry from bad inputs. Structure-from-motion will happily reconstruct a plausible fastener that does not exist. Scale must come from a measured reference in frame, not from the solver.

What this framework cannot do, in its current state

1. It cannot show you a true section of either bridge. Both drawings here are schematic and not to scale. For the Williamsburg Bridge exactly two transverse dimensions are stated in any located source — the 67 ft stiffening truss width and the approximately 40 ft truss depth — and neither locates any element within the section. For the Manhattan Bridge none are, at better than a tertiary source. Every one of the 27 components is placed by reasoning.
2. It cannot resolve the shielding hypothesis. Whether the Williamsburg Bridge's flanking orthotropic roadway decks actually interrupt the path from rail to a receptor below depends on the relative vertical positions of the track, the deck and the truss bottom chord. That is precisely what no located source states. A deck rigidly connected to the floor system may also radiate rather than shield, which geometry alone cannot settle either.
3. It has no fastener-level content, because no public source has any — and geometry would not be enough anyway. The brief asks for schematics down to screws. The rail fastening specification for both bridges is a records request, not a modelling problem. Even a complete fastener geometry would not supply dynamic stiffness, loss factor or clamping preload, which are the parameters a vibro-acoustic model actually consumes and which come from a specification or a test.
4. It is not an acoustic model and should not be dressed as one. No sound propagation is computed anywhere in this artifact. The sightline overlay is a straight line, which is a statement about geometry and not about sound.
5. Its element colours describe geometric support, not source quality. Two components are drawn from a fully verified source and still marked assumed, because the source establishes that the element exists without locating it. Reading the amber and red as "bad sources" is the specific misreading this schema was built to prevent.
6. It encodes one author's inferences. The provenance states are assigned by the same person who drew the geometry. That is a conflict, and the mitigation is that every inference is labelled and every source is named, so a reviewer can attack them individually.