Getting started

Quickstart

This walkthrough covers the two things most people do first: check a claim, and pull a synthetic sample. Both run entirely from a local install — no account or API key is required for either the Python client or the public source registry.

Install

pip install equilibrium-labs

See installation if you need the MCP server extra or want to pin a specific version.

Validate a claim

claimcheck.validate takes a plain-language quantitative claim and returns a structured result: the sources behind it, the assumptions required to connect those sources to the number, and a confidence band.

from equilibrium import claimcheck

r = claimcheck.validate("AI raised task exposure 14% in 2025")

print(r.confidence_band)
# (0.09, 0.19)

for source in r.evidence:
    print(source.citation, source.retrieved)

The point estimate (14%) sits inside the band, but the band — not the point estimate — is the thing to report. See validating a claim for the full shape of the result object.

Generate a synthetic sample

The household panel generator produces synthetic respondent-level data calibrated to EU-SILC moments. Every run is deterministic given a generator name, a sample size, and a seed.

from equilibrium import synthetic

sample = synthetic.generate("household_panel", n=2_000, seed=20260801)

sample.shape
# (2000, 34)

sample[["household_size", "disposable_income", "region"]].head()

Run it again with the same arguments and you get byte-identical output — the seed, the calibration target version, and the generator version are all recorded in the run’s provenance manifest.

Where to go next

• Generators for how the generator architecture works
• Validating a claim for the full claimcheck result object
• MCP server overview to expose both as LLM tool calls