Getting started

Overview

Equilibrium Labs builds infrastructure for economic research that has to survive scrutiny. We are independent (funded without institutional strings attached, the ledger kept anonymous), open (code under MIT, data and papers under CC BY 4.0), and reproducible (every figure we publish ships with the notebook, the seed, and the environment lockfile that produced it). The toolkit is built in Denmark and used by people who need to know exactly where a number came from.

Three pillars

Synthetic data. Generators produce household and economic panel data calibrated to real-world moments — income deciles, household size distributions, employment shares — drawn from sources like EU-SILC and Eurostat. A generator is fully specified by its calibration target, its sampling procedure, and a seed; run it twice with the same three inputs and you get the same output.

claimcheck. A validation layer for quantitative economic claims. Feed it a sentence like “AI raised task exposure 14% in 2025” and it returns the sources the claim rests on, the assumptions required to get from those sources to that number, and an uncertainty band around the estimate. Nothing is asserted without provenance.

The Equilibrium MCP server. Wires the generators and claimcheck into a model context, so an LLM can call validate_claim or generate_synthetic_sample as tools rather than guessing from training data. Every tool response carries citations and a confidence band by design — this is not an optional flag.

Who this is for

Applied researchers who need synthetic panels that reproduce known moments without exposing real respondent data. Journalists and analysts fact-checking a quantitative claim in a policy brief or news story before repeating it. Model builders who want their LLM’s economic claims grounded in a source and a number, not a plausible-sounding sentence.

If you’re getting started, the installation and quickstart pages are the fastest way into the package.