Thesis Institute
We build agents whose job is to predict consequential public facts, explain the evidence, call inspectable tools, and learn from scored outcomes.
The stack
Open forecasts
The public surface tracks government statistics, policy settings, and conditional outcomes with resolution rules and calibrated uncertainty.
PolicyEngine
Tax and benefit forecasts call open microsimulation models instead of treating model outputs as hidden oracle claims.
Microplex
Calibrated microdata lets agents test policy scenarios against transparent population structure and administrative benchmarks.
Brier Decisions
The personal decision package keeps the same discipline for local choices: outcomes, uncertainty, review dates, and calibration.
Forecasting as alignment pressure
Forecasting gives AI systems a narrow public job, a hard feedback loop, and a record others can inspect. The work compounds when each resolved outcome improves the next agent run.
Define the public outcome and the exact source that will resolve it.
Call official data, encoded law, PolicyEngine, and other inspectable tools.
Publish the forecast distribution, public trace, and later score.
Public preview