“AI grading” can mean a black box that spits out a number. That’s not what we built. On Gradogram, a score is only ever the start of a conversation — the criteria are yours, the reasoning is visible, and a human makes the call. This post walks through the whole path from a candidate’s answer to a confirmed decision.
1. You define the rubric
Before a single candidate answers, you set the criteria that count for the role. The rubric is frozen for the run, so every candidate is graded against the same explicit standard rather than a shifting sense of a “good answer.” Consistency is the whole point: the same criteria, applied the same way, every time. For a full walkthrough of what makes a criterion gradable in the first place, see how to write a hiring rubric that's actually gradable.
2. The AI grades against it
When an answer comes in, the AI grades it criterion by criterion. For each one it decides whether the criterion was met, partially met, or not shown, and it records the evidence it drew from. It aggregates those into an overall score and writes a plain-language synthesis with a recommendation.
- A per-criterion breakdown — met, partial, or not shown — you can check against the answer.
- An overall score with a model confidence read, so you know how sure the model is.
- A short written explanation of why, in language a hiring manager can follow.
- Anything the AI can’t confidently grade is flagged for manual review, not failed silently.
3. A person confirms the decision
The score is decision support, not the decision. A member of your team reviews the reasoning and confirms it before it counts. The platform is designed so that the consequential call — the one that affects a candidate — stays with a person.
Every score is recorded, immutably
Each grade, override, and confirmation is written to an immutable audit trail — who, what, and when. AI-generated content is clearly labelled as such wherever it appears. If you ever need to explain how a decision was reached, the record is there.
Why the audit trail matters
Automated help in hiring only earns trust if it can be inspected. Because the rubric is explicit, the reasoning is shown, and every action is logged, you can answer a simple question at any point: why did this candidate get this score? That traceability is built in, not bolted on. Candidate data can also be anonymized on request to support right-to-erasure needs, and each organization’s data is isolated by a framework-enforced tenant boundary.
This same flow runs under all four assessment types — Knowledge, System Design, Behavioral, and Culture Fit. Different formats, one grading engine, one principle: AI grades, you decide.