§About

Equality before the law, applied to the models that decide.

isonomy/aɪˈsɒnəmi/ · noun
  1. From Greek isonomia: equality of political rights; equality before the law.
  2. The standard we hold automated decisions to: the same rules, applied the same way, to every group, and every intersection of groups.

§ 01Why we exist

Hiring is moving to models faster than audits are moving with it.

Most bias audits today report one attribute at a time, once a year, and stop at the number. That misses the people at the intersections, the inputs that act as proxies, and the release that quietly made things worse.

Isonomy exists to close that gap with methods from current fairness research, built so a vendor’s engineers, an employer’s counsel and an independent auditor can all read the same evidence.

§ 02Founder

Gideon Popoola, Ph.D.

Gideon’s doctoral research in Computer Science at Montana State University studies how unfairness hides in machine learning: across intersectional subgroups, inside model representations, and in the explanations models give.

Before founding Isonomy he built machine-learning systems in data science roles at Walmart (delivery forecasting), Riot Games (player-behaviour risk) and JPMorgan Chase (credit-default modelling and drift monitoring). He started in data science in Nigeria and South Africa, building education and health analytics.

An audit that stops at one attribute and one number misses where the harm usually sits: at the intersections, and in how the model reasons.

§ 03Research

Peer-reviewed foundations.

Isonomy’s intersectional, procedural and explanation-fairness analyses build on this work by Popoola and Sheppard:

  • MESD: A Risk-Sensitive Metric for Explanation Fairness Across Intersectional SubgroupsIEEE Symposium Series on Computational Intelligence (SSCI), 2027 (accepted)
  • Procedural Fairness via Group Counterfactual ExplanationECML PKDD, 2026
  • Beyond Equal Outcomes FairnessIEEE Conference on Artificial Intelligence (CAI), 2026
  • Investigating and Mitigating the Performance–Fairness Tradeoff via Protected-Category SamplingElectronics 13(15): 3024, 2024 · doi:10.3390/electronics13153024

§ 04Principles

What we won’t do.

  • Recommend group-specific cut-offs or score adjustments. They are unlawful for employment tests, and they hide problems rather than fixing them.
  • Treat passing the four-fifths rule as proof that a tool is fair.
  • Build or sell hiring models. We only evaluate them, so our findings aren’t shaped by a product of our own.
  • Sell, share or reuse the data you send us for anything but your audit.
  • Hide our method. Every statistic we report is documented, so it can be checked.

Talk to the person who runs your audit.

At this stage you work directly with Gideon, from the first call to the findings walkthrough.