Pseudo-PFLOW: making the freight that surveys miss

Status: in progress, CSIS / Sekimoto Lab, The University of Tokyo. Project card: Pseudo-PFLOW. This post is the question and the setup. It is not a results paper.

Why it exists

Household travel surveys are where I learned the craft. They are also almost silent on business vehicles. Trucks and taxis eat urban road space and they barely appear in a diary of “where did you go yesterday.” If you only calibrate a city on households, you are modelling the commute and calling it traffic.

Pseudo-PFLOW is a nationwide agent-based picture of those missing vehicles: synthetic trucks and taxis, daily activity, assigned to a zone system, checked against statistics that are published. The CV numbers I will repeat, and not add to:

• on the order of 4.5 million daily trips
• 1.5 million-plus truck and taxi agents
• 106 zones
• calibration and cross-verification against source statistics (not against a private GPS dump I will not name here)

I also wrote technical documentation for the truck and taxi ABMs: calibration, data provenance, cross-verification. That sentence is on the CV because provenance is the method. If you cannot say which table an agent’s trip rate came from, you have a simulator, not a synthetic population.

What the model is for

A planner who wants national freight and taxi load on the network without waiting for a new establishment survey. A researcher who wants a baseline of business-vehicle kilometres that a passenger-only four-step will not give them. Later, a check on whether a generative mobility model is inventing truck-shaped trips or only commuting.

It is not a digital twin of every depot. It is not GUFM. GUFM is the other Tokyo thread and it is not this series.

How I am willing to talk about progress

Same rules as the method post:

• The unit is the agent-day, the geography is the zone, the judge is a published margin (counts, totals, shares) — not a loss curve.
• If a calibration gap is not in a public figure, it is not on this blog.
• Code and papers, when they exist as citable objects, go on publications and GitHub. Until then, the project page is the placeholder.

When there is a paper, this post gets a DOI line. Until then: 4.5 million trips, 106 zones, and a refusal to fake the fit.

The layer after a synthetic population is action-conditioned dynamics: world models for mobility.