Subsurface World Models (SWMs) are a new class of physics-grounded, uncertainty-calibrated AI designed to infer, predict, and reason about the Earth's hidden interior from sparse, indirect measurements. Unlike world models of scenes a camera can see, the subsurface is never directly observed and its ground truth is never revealed — so SWMs pair a learned geological prior with known physics to deliver not a single image but a calibrated distribution over the earths consistent with the data.
At field scale, across exploration-and-production (E&P) activities, an SWM becomes a reliable, continually updated environment for engineering practice — a place to steer wells, forecast how the reservoir responds to drilling and injection, and choose the next action by the uncertainty it removes rather than by a single best guess.
The notes, the benchmark and the engineering side have moved to one place.
This page is the summary. The research program in full, the eight notes, the glossary and reading list, and the engineering agents that use the model as their environment are at subsurfacelab.no, shared with registered readers.
If you think this is wrong, or you have a better idea, that is the most useful email to send.
Especially if you work on inverse problems, generative models for scientific data, subsurface characterization, blind-well datasets, or calibration benchmarks. Notes and results will be published here as they come.