Noto Peninsula · Jan 1 2024 · Mw 7.5
how the model
learns to think
we build a tiny town of survivors out of a real evacuation survey — then watch them decide, in words, as the water rises.
1
a model of the town
tuned until it matches the real Noto survey.
2
a resident inside it
a real survey role — a house, a memory, a worry.
3
the Lens
the engine turns raw data into what they see, in words.
4
the Hand
their choice in words, written back as a move on the map.
a calibrated agent-based model, a population of LLM residents, and an Environment Engine that translates both ways — built on real GIS, seismic, and warning data from the 2024 Noto earthquake & tsunami.