Hiring fit-scores that don't play favorites

Hiring software has a trust problem, and it's earned. The typical "AI resume screener" is a black box: the same candidate can score differently on different days, nobody can explain why, and the vendor's answer to "how does it work" is a shrug in a sales deck. Then companies use those scores to filter human beings out of jobs.

That is a bad product and, increasingly, a legal liability.

What deterministic means

Deterministic scoring is simple: the same inputs always produce the same score. Run a candidate through twice, get the same number twice. It sounds basic because it is basic — and most hiring tools can't do it, because their models are probabilistic systems with no reproducibility guarantee.

Determinism matters for one reason: defensibility. If a candidate asks why they were scored the way they were — or a regulator does — "the model felt like it that day" is not an answer. A reproducible score with a stated method is.

What versioned means

Models change. When they do, most vendors silently rescore your pipeline — yesterday's 82 becomes today's 74 and nobody tells you. Versioned scoring ties every score to the model version that produced it. Old scores keep their old version label. Nothing is rewritten by a model the candidate never met.

This is how you avoid the quiet scandal: a candidate rejected under version 3 who would have passed under version 2, with no record that the rules changed mid-game.

Why advisory-only is the whole point

Here's the line most vendors won't draw: the tool must not make the hiring decision. Advisory-only means the engine scores and explains, and a human being — with a name — advances or rejects. The tool cannot auto-reject. It cannot auto-advance. It has no button for that.

This isn't humility; it's liability management. Automated employment decisions are where hiring software meets employment law, and the companies that let the tool decide are the ones that end up in the news. The score is evidence for a human decision, not a decision with a human rubber stamp.

The honest trade-offs

• Scores don't hire people. A deterministic score is a better input, not an answer. If your hiring process is broken — bad criteria, biased interviewers, no structure — a good score won't fix it. It'll just measure the brokenness reproducibly.
• Deterministic isn't unbiased. The scoring rules are written by humans, and humans have biases. Determinism guarantees consistency, not fairness. The advantage is that consistent rules can be audited — you can actually check a deterministic system for bias, which is more than you can say for a black box.
• Someone still has to decide. Advisory-only means the work of deciding stays with you. That's the point, but it's worth saying plainly: this tool will not shrink your hiring workload to zero. It shrinks the arbitrariness.

The hiring tools worth trusting are the ones that show their work, version their changes, and refuse to decide for you. Everything else is a black box with a pricing page.