Candidate screening software that ranks on evidence

Years of experience correlates with job performance at close to zero once you are past the first year or two. Educational prestige correlates with the socioeconomic status of the candidate's parents at least as strongly as with anything you want. Both are legible, both are fast to filter on, and both encode exactly the demographic history you are legally obliged to avoid replicating.

Automating that filter does not remove the bias. It launders it, and it produces a defence that is worse in litigation than a human decision, because the pattern is now systematic and documented.

Candidates complete a fifteen-minute adaptive psychometric battery. Scores are placed against the benchmark for the relevant job family, and a role-fit estimate is computed from the trait-to-demand mapping for that role — not a global 'good candidate' score, because no such thing exists.

Rankings are explainable by construction. Clicking any candidate's position shows the trait scores, the benchmark, the validity flags and the weighting that produced it. Where a profile carries an impression-management flag, the candidate is surfaced as unreadable rather than quietly ranked low.

Selection rates by protected group are computed continuously against the 4/5ths rule from the Uniform Guidelines. If your pipeline drifts, you find out while you can still act, rather than in an audit.

New York City's Local Law 144 requires an annual independent bias audit of automated employment decision tools and public disclosure of the results. The EU AI Act classifies employment-related AI as high-risk. Both regimes assume you can produce per-decision records. TalentSpark logs them because you will eventually be asked.