Ask a recruiter why a candidate scored an 84 in most AI screening tools and you'll get a shrug. The number came from somewhere, but nobody can point to where, and the honest answer is usually "the model thought so." That's a black box, and it leaves you two bad options: follow the number blindly or ignore it. Neither is something you can defend to a hiring manager.

We built SquarePeg on a different bet. It's AI-assisted, not AI-automated. The score ranks your candidates and explains itself, and a person on your team still decides who moves forward and who doesn't. That's what we mean by glass-box scoring, and it's worth walking through exactly how it works, because "explainable AI" gets used as a marketing phrase a lot more often than it gets used honestly.

The score sets your reading order. It never makes the call.

Start with what SquarePeg doesn't do, because it's the part that matters most. There's no auto-reject below a threshold and no auto-advance above one. SquarePeg doesn't move a candidate anywhere; advancing or rejecting someone is always an action a recruiter takes. That's a deliberate product decision, not a gap on the roadmap. The problem the score solves is narrower and more honest: nobody can read every resume for every req closely, so something has to decide what gets read first. The score does that, and then explains itself well enough that you can disagree with it.

No mystery number. The total is built from the criteria you wrote.

When you set up a job in SquarePeg, you write criteria: the qualifications, experience, and screening answers you're actually looking for. Each one is scored individually, 0 to 100, against the candidate's enriched resume, their company data, and any relevant screening answers. The total score is a weighted average of those individual scores, with nothing hidden on top. There's no separate step where a model looks at the whole picture and produces a number with no path back to what created it.

You decide what matters.

Every criterion carries a weight you choose. Required criteria count 3x: the core qualifications that move the total the most. Preferred criteria count 2x: the things that help you tell strong candidates apart. Bonus criteria count 1x: nice-to-haves that nudge a score without dominating it. You set the weight when you write the criterion, so the ranking reflects your judgment about this specific role, not a generic idea of a "good candidate" baked into a model somewhere.

Every score shows its work.

Open a candidate's profile and the Score section doesn't just list numbers next to labels. Each criterion has a plain-language explanation of how its score was determined: what in the resume, the enriched company data, or a screening answer produced it. If a candidate scored well on "five-plus years in a regulated industry," the explanation tells you what in their history counted as evidence, not just that they passed.

Then you can go one step further. Click into a criterion and use the arrows to step through the exact resume text, company data, or screening answer behind it. This is the part that turns "trust the algorithm" into "check the algorithm's work." You're looking at the same evidence the score used, not taking our word for it.

Disagree? Fix the cause, don't just override it.

Disagreement should be useful information, not something you quietly work around. If a score is off because the wording was too broad, tighten the criterion. The tool did what you asked, but what you asked wasn't precise enough. If the score is right but shouldn't count as much, change the weight. If SquarePeg misread what you meant, say by misunderstanding an acronym or missing context a person would catch instantly, use the Feedback button in the Score section. We review that feedback regularly to improve the scoring algorithm and the company enrichment behind it. And when we ship an algorithm update, you'll see a banner and a Rescore Candidates button, so you can rescore your existing candidates against the latest version whenever you're ready.

Why a glass box matters more as volume goes up, not less.

The hard part of recruiting right now isn't ranking. It's that there's more to read than anyone can read. If the thing sorting that pile is a black box, you're trusting a number you can't inspect at exactly the moment, high volume and high stakes, when you most need to catch a bad call. Glass-box scoring doesn't try to replace a recruiter's judgment. It makes sure that judgment still gets exercised, on a shortlist that's already sorted and explained well enough to argue with.

That's the test we hold it to: can a recruiter look at a score, disagree with part of it, and say specifically why? If they can't, the explanation isn't doing its job. If they can, we've done ours.