Sim command — logged and replayed
These are the run. Each is written to the command log, so the same network,
seed and log replays identically — which is what makes a share link a run rather
than a screenshot.
Signal
Hold and call change the run, and they replay with it.
Readouts
Approaches
Queue is instantaneous. Delay is a running mean per vehicle, and it is
uncalibrated — see About.
Spec change — costs you the run
An assumption is not a setting: it is part of the scenario, so changing one starts
a new run from t = 0 rather than steering the one you are watching. There
is no way to make that cheaper — the worker's only entry point for a scenario is
init — so the panel says so instead of hiding it.
This run
A run in this model is a network, a seed, and the list of commands issued to it —
nothing else. Pause, speed, step and camera cannot reach the simulation
(D11 , and it is a physical boundary rather than a convention: the command
type has no case that could carry them), so every edit is already classified as
shareable or not, continuously.
That is why the note's worry about needing an extra click to "lock in" a shareable
URL does not apply. Every edit is already in the log, so the button below reads what
the run has actually been told to do rather than asking you to declare it first.
The link carries the network, the seed, your assumption edits and the command log,
compressed into everything after the # — which never leaves this
browser, so a run you looked at is not a run anybody logged. Opening one replays it
from the start at normal speed and the commands fire where they fired.
Free, today, for everyone
Every network in the picker, at any demand, with the aerial backdrop
Pedestrian delay, queues, approach delay, and the integrity readouts
No account, no sign-in, and nothing stored on a server — there is no server
What a paid tier would have to earn first
Charging for "modelling ability and analysis reporting" means charging for delay
and level-of-service figures, over a named intersection, with a PE's name on them.
Two calibration gates are still open — saturation flow and the free-flow speed
distribution — so those numbers are shaped correctly and sized arbitrarily.
A free surface may carry figures labelled uncalibrated. A paid one may not. So the
paid axis, when it arrives, is machinery rather than numbers: replication with
confidence bands, saved projects, export, larger areas.
There is no waitlist form yet, and no payment portal. Mail me instead.
What this is
A microscopic traffic simulator that runs entirely in your browser — car
following, gap acceptance, signal control and pedestrians, stepped ten times a
second. Nothing is uploaded, because there is nothing to upload to.
What it cannot tell you yet
Delay and level-of-service figures are uncalibrated. The model
reproduces the right shapes and the sizes are not yet pinned to observation:
saturation flow needs a signalised standing queue, which Braddock & Motor does not
have, and the free-flow speed fit is closed on that site's speeds only. Treat every
delay figure as a comparison between two runs of this model, never as a prediction
of a real one.
Three unmeasured behavioural assumptions move the pedestrian results by more than
the model does: the share of drivers who yield to somebody waiting at a kerb,
whether people cross 18 m in one stage or two, and the corner return radius. They
are reported as ranges rather than as numbers, and no before/after has been sealed.
Where the geometry comes from
Braddock & Motor is anchored on a Census reference point, its street centrelines
are fitted to OpenStreetMap, and its volumes are an October 2025 tube-and-camera
collection. The corner radii, turn ratios and yielding behaviour are not measured.
The aerial is context for the reader and nothing is measured off it.
Street geometry © OpenStreetMap contributors, ODbL. Imagery © Esri and its
partners, with a public-domain USGS fallback.
Contact
Contact address to be set before launch.