For employers reading a candidate's scored interview.

​

Truffle scores every completed interview on a 0–100 scale, calibrated to the specific role you're hiring for. Each question gets its own score, backed by a rubric built from your job description and intake answers. Every score includes callouts — short notes pinned to specific moments in the candidate's answer — and an executive summary that points at where to probe in the next round.

​

When you lock in a role, Truffle generates a role-specific rubric for each interview question. The same question — "Tell me about a time you handled a difficult customer" — produces different criteria for a Customer Success Manager than for a Sales Ops Analyst. The CSM rubric weights empathy and de-escalation; the Sales Ops rubric weights data-driven follow-through.

​

The score reflects what you care about for this role, not a one-size-fits-all standard. Better intake produces sharper scoring. See How do I personalize Truffle's AI for my position?.

​

When you set up a new role, Truffle recommends 5 interview questions from a curated bank, chosen to fit the role's seniority, function, and the context you provided in intake. You can swap any of them, add your own, or remove ones that don't fit. The recommended set is a starting point, not a prescription.

​

Every scored interview has a short executive summary at the top of the review page. 30 seconds. It tells you whether the candidate is worth a closer look, what stood out across the interview, and where to probe — the specific questions or criteria where the candidate's answer left an open thread worth chasing in a live round.

​

If the summary doesn't make you want to dig in, you usually don't need to. If it does, the "where to probe" section is your agenda for the human interview.

​

A topline score of 68 can hide a 90 on evidence of impact and a 25 on communication clarity. The gap is often more informative than the average.

​

Click into any scored question to see the per-criterion breakdown. Each criterion shows what the rubric was looking for, what the answer provided, and where it fell short.

​

Each scored question has callouts — short notes pinned to specific moments in the candidate's answer. A callout might flag a strong example the candidate offered, a place where the answer fell short of what the rubric was looking for, or something worth a recruiter's second look.

​

Callouts are the AI's "show your work" view. If you want to know why a question scored a 70 instead of an 85, the callouts usually have the answer — and they often point at exact phrases from the candidate's response.

​

Reach for callouts when:

​

​

If the score doesn't match your read of the answer, the rubric explains what the AI was looking for. Sometimes that surfaces a real miss in the scoring; more often it reveals that the rubric weighted something differently than you would have — which is useful information for calibrating intake on the next role.

​

A 0 on a question means nothing scorable in the response — the candidate dodged, answered something else, or there was a technical issue that broke the recording. Treat any 0 as "investigate before deciding." Common causes:

​

​

Sort candidates by topline score to prioritize who to look at first. Let the actual advance / reject call stay with you. The score is signal, not a verdict.

​

Here's a concrete example. You're hiring for a Customer Success role. A candidate scores 56 overall — looks like a pass at first glance. But the breakdown shows:

​

​

Without the breakdown, you'd cut this candidate. With it, you bring them into the human round and ask the tooling question live. That's a candidate you would have lost to a topline number.

​

The breakdown — combined with the callouts on each question and the "where to probe" line in the executive summary — tells you exactly what to ask in the human round. You go in with a sharper agenda instead of re-running the same questions the AI already covered.

​

An 84 on a Sales Ops role and an 84 on a Therapy role are scored against different rubrics. They are not the same bar. Compare candidates within a role's pipeline; don't benchmark across roles.

​

If your read of a candidate and the score disagree, your read wins. But before you move on, look at the rubric: what was the AI weighting that you weren't, or what did you see that the rubric missed? Either answer is useful — it either calibrates your intake for next time, or it tells us something about the scoring we should know.

​

Truffle's scoring is tested against systematic fairness checks before any model change ships. We run candidate answers through prompt-level variants that probe for patterns that could disadvantage candidates by background, communication style, or accent — and we don't ship a change if those variants move scores in ways they shouldn't.

​

We also produce audit artifacts in the format NYC's Local Law 144 expects, which our employer customers can hand to a third-party auditor when required.

​

​