How does RChilli test models for bias across gender, ethnicity, location, age, or other sensitive attributes?

RChilli uses multiple testing methods to evaluate fairness.

Counterfactual testing

In counterfactual testing, two or more versions of the same profile are created with identical professional qualifications, while one potentially sensitive attribute is changed. Examples include changing the name, graduation year, or pronouns. The system then compares match score, ranking position, and top-N visibility. The goal is to confirm that non-job-relevant personal changes do not alter the outcome unless there is a legitimate job-related reason.

Proxy bias testing

RChilli also tests for indirect bias. Even when sensitive attributes are excluded, other variables may still act as proxies. Examples include education background, career gaps, and work history patterns. These are reviewed to ensure they are not creating unfair advantages or disadvantages without a clear job-related basis.

Ranking fairness analysis

Because hiring systems often rely heavily on top-ranked results, RChilli evaluates whether candidate ordering is fair and consistent. The analysis checks Top 5 and Top 10 visibility, whether equally qualified candidates receive similar ranking positions, and whether hidden preference signals are affecting exposure.

Together, these methods help validate that direct and indirect bias are both being controlled.

If you need further assistance, feel free to contact the RChilli Support Team by sending an email to support@rchilli.com.