What the measured levels actually sound like

Why continuous running is the more honest mode

A single pass-by is an event. What a person sitting in the park experiences is a process: the same event, on a headway, for twenty hours a day. Those are different claims about harm, and only the second one is the problem the MTA was asked about. The headway is published — it is the session duration divided by the train count, and at the dog run that is 26 trains in 37 minutes 45 seconds, one every 87 seconds — so the process can be modelled from data just as directly as the event can.

Three things change when the loop runs rather than a single pass.

1. The gaps become the argument. At the dog run the train is meaningfully above background for roughly forty seconds out of every eighty-seven. Long enough to notice; not long enough to adapt to. Written as a duty cycle — 7.2% of the session at full peak-equivalent energy — it sounds small. Sat through, it is not.
2. The energy average becomes measurable on screen. The running Leq readout accumulates the modelled power over modelled time and converges. That is the number CEQR, the FTA and § 1204-a are all written in, and it is the number the MTA published. Watching the loop settle onto it is the closest this page comes to checking itself.
3. The tails overlap. The loop sums the three nearest events rather than playing one at a time, because on a four-track crossing the tail of a departing train is still arriving when the next one begins. Playing isolated events back to back would quietly discard that energy.

On time compression. Real time is offered and is the default, but a real headway means real waiting. The compressed rates scale every interval by the same factor — the gaps and the pass-bys alike — so the duty cycle is preserved exactly and the energy average is unchanged. The only thing compression falsifies is the tempo: the trains sound faster than they are. Levels, differences and averages all remain what the measurements say.

Two receptors have no continuous mode, on purpose. Front and Pine Street and the Williamsburg Bridge walkway publish no train count and no session length, so neither has a measured headway. Running a loop there would mean inventing the service pattern, and the button is disabled rather than filled with a plausible guess.

What is measured, what is derived, and what is invented

Every parameter of this demonstration carries its own provenance, on the same DOCUMENTED / INFERRED / ASSUMED scale the rest of this programme uses, with one addition: DERIVED means it was calculated from published numbers by arithmetic shown in full below.

The one thing this demo derives that the sources do not state

The MTA publishes, for each public-space session, four numbers: the session Leq, the peak Lmax, the train-free baseline, and the number of trains over a stated duration. It does not publish how long a train event lasts. But those four numbers determine it, if you assume the session is baseline plus N equal train events. Solving the energy balance for the equivalent time each train would have to spend at its peak level gives:

Thresholds this sits against

From the CEQR Technical Manual noise-exposure guidelines, as cited in IDEA-CONCEPT.md §3.2. These are L10 criteria; the MTA reported Leq, so a direct comparison is indicative and not a formal classification.

Where this demonstration is likely to be wrong

1. The spectrum is fabricated. No third-octave data has been published for either bridge. Two sounds can share an A-weighted level and be entirely different experiences. Nothing here should be used to argue about annoyance, speech interference, or which frequencies a treatment should target.
2. The envelope shape is a convention, not a measurement. The decay law used is a compromise between point-source and line-source behaviour. Only its energy is pinned to data; its shape is chosen.
3. The derived event duration assumes all trains are equal. They are not. B, D, N and Q services differ in car class, consist length and axle count, and the MTA report normalises for none of these. The derived figure is an average over an unknown mixture.
4. Continuous running assumes a perfectly regular headway. Real service is not a metronome. The published figure is a session average — 26 trains in 37 minutes 45 seconds — and the real intervals scatter around it, bunch behind delays, and differ by direction and by service. The loop is a mean rendered as a rhythm, and the regularity you hear is an artifact of that. It also runs a single headway indefinitely, whereas real service thins overnight; the MTA sessions were daytime, so the loop represents a daytime hour and not a day.
5. It assumes the published baseline is a true train-free floor. At the DUMBO Archway the "baseline" is 68.9 dB(A), which is not a quiet environment. Some of what is attributed to background may be distant rail.
6. The meter agreement is engineered, not evidence. The synthesis is scaled so the rendered A-weighted level matches the modelled level. That the two agree confirms the scaling works. It confirms nothing about the bridge. Under time compression the meter's averaging window is shortened in proportion so that it still spans a comparable slice of modelled time; that keeps the check fair but makes the rendered needle visibly noisier in the gaps, where it is averaging a shorter piece of a random signal.
7. So is the running-average convergence. The energy average of the loop settles onto the published Leq because the event duration was solved from that Leq. It is a closed loop and it proves the arithmetic is consistent, not that the model is true. It would be evidence only if the duration had come from somewhere else. It is also not exact. The derivation in §1.7 treats a train event as replacing the baseline for its duration, whereas the loop adds the event on top of a baseline that never stops, so the loop carries slightly more energy than the published figure: 0.002 dB more at the dog run, 0.021 dB at the DUMBO Archway. The residual is stated rather than removed, because removing it would mean editing a published derivation to make an audio demonstration look tidier.
8. Peak levels are single events. A Max Lmax is the loudest thing that happened once. Treating it as the level of every train overstates the typical case, which is why the derived durations below are labelled "equivalent" and not "actual".
9. The indoor sites break the derivation and are shown here precisely so the failure is visible rather than hidden.

A correction to the first version of this page

The version of this artifact published on 1 August 2026 carried a button labelled "Continuous, at the measured headway". It was not. The code capped the interval between trains at thirty seconds:

That is withdrawn. No receptor on this page has a thirty-second headway; the measured figures are 85 to 126 seconds. The cap inflated the duty cycle by between three and four times, which would have driven the energy average several decibels above the published one, and the label asserted a fidelity the implementation did not have. It is recorded here rather than quietly fixed because that is this programme's rule, and because it is a good example of the failure it keeps having to correct: a plausible-sounding claim attached to something that was never checked against the number it named. The running-average readout now on the page exists partly so that this class of error cannot recur silently — a capped headway would show up immediately as a running Leq that refused to settle on the published figure.