Open infrastructure for honest economic analysis.

Equilibrium Labs exists because AI systems are now doing a large and growing share of applied economic work, and they are doing it without the practice that makes such work trustworthy. A model will produce a fluent regression table whether or not the design identifies anything, report a contested empirical question in the voice of arithmetic, and compare two series that do not measure the same thing. The output looks the same in every case, which is what makes it dangerous.

The project's answer is not a product but a set of standards, published in full and written in a form an agent can be held to. Everything is open source — MIT for code, CC BY 4.0 for the written material — with no paid tier and no hosted edition. That is a deliberate commitment rather than a stage: a standard that is partly withheld is not a standard, and guidance that shapes analysis other people publish has to be readable, criticisable, and forkable by anyone who disagrees with it.

The lab is built by Alfred Birkelund, a university student in Denmark who came to economics from engineering. It grew out of a practical worry rather than an academic one: agentic systems are increasingly asked to weigh economic ideas and policy proposals, and they do it carrying priors they never declare. The aim is to make that more robust — to give professionals tools that are faster without being looser, and to put the correction for a model's inherited biases somewhere it actually operates: in the skills a harness loads at the moment the analysis is being done, rather than in a disclaimer appended afterwards.

A concrete case, and the one that motivated much of this: think tanks and institutes publish policy proposals continuously — some avowedly aligned with a political tradition, others closer to plain applied economics. Being able to put any of them through the same method, without the evaluation quietly bending to whose proposal it is, is the capability this is meant to produce. That is a statement about method, not a position on any proposal. The project works with anyone and endorses no one.

It is maintained by one person at low intensity. There is no support commitment and no roadmap obligation, and both are stated plainly so nobody plans around something that does not exist. What is committed is that nothing published here will be relicensed, paywalled, or withdrawn.