Scoring Criteria

Every Recruiter search has a scorecard — a small set of criteria that describe what makes a strong candidate for the role. Recruiter generates the scorecard for you, then uses an LLM to score every fetched candidate against each criterion and roll those into a single weighted aggregate score. That score is what ranks the candidate list, and the per-criterion breakdown is what you read to understand why a candidate ranked where they did.

Beta. Recruiter is in beta. Scorecard generation, the scoring scale, and weighting may change.

What a scorecard is

A scorecard is a set of criteria. Recruiter generates exactly five criteria per search, derived from your query, your filters, and your job description (if you uploaded one). Each criterion has:

A heading — a single word that labels the criterion in the candidate table (for example, "Cloud," "Leadership," "Scaling").
A description — one short, specific sentence describing the evidence the criterion looks for.
A type — primary or secondary.

Type	Meaning
Primary	A must-have for the role. Weighted more heavily in the aggregate, so a candidate who misses it scores low.
Secondary	A nice-to-have. Weighted less, so it boosts the score when present without heavily penalizing its absence.

A realistic scorecard mixes both — typically two to three primary and two to three secondary criteria. The scorecard populates even when you didn't upload a JD; it's built from whatever you provided when describing the role.

How the scorecard is generated

When you submit a search query (or regenerate the scorecard), Recruiter makes a single LLM call that produces the five criteria. The generator is given your query, a human-readable summary of your active filters, and your JD text (truncated to roughly the first 4,000 characters). It's instructed to write criteria that are specific, evidence-based, and assessable from a professional profile — so the criteria favor concrete, observable experience over generic traits, and they're written to be readable by a non-technical recruiter.

A few properties are baked into generation:

Exactly five criteria, no more and no fewer.
Each criterion targets a different dimension of the role, so two criteria don't probe the same gap.
Criteria describe a spectrum of depth (from shallow to deep experience) rather than a yes/no checkbox, so experienced candidates spread out across scores instead of all landing at the top.
The single most critical responsibility is marked primary; the rest are balanced between primary and secondary.

You can edit the generated scorecard before — or after — running the search: rename or reword any criterion, change its type, and add or delete criteria. When you add or edit a criterion's text, Recruiter regenerates its one-word heading in the background, so the heading may show as pending for a moment after an edit.

How candidates are scored

Once candidates are fetched, Recruiter scores them against the scorecard with an LLM. Scoring runs as a background job — one per search — and processes candidates in batches, so the candidate list shows scores filling in over time rather than all at once. A candidate whose scoring hasn't completed yet shows a generating state in the score column; the page polls and updates as results land.

The scoring scale

The LLM scores each criterion on a 0–5 integer scale:

Score	Evidence level
0	No evidence in the profile.
1	Speculative.
2	Weak.
3	Partial.
4	Strong.
5	Multi-context — demonstrated across more than one setting.

The model must return a whole-number score for every criterion on every candidate — when evidence is thin or absent it returns a low score with a short reason rather than skipping the criterion — and each score comes with a one-to-two-sentence justification that cites specific evidence from the profile. That justification is what you read in a candidate's scorecard to see the reasoning behind a number.

The aggregate score

A candidate's overall score is a weighted average of their per-criterion scores. Primary criteria are weighted more heavily than secondary; the default weights are 0.6 for primary and 0.4 for secondary. The aggregate is computed only over the criteria that were actually scored for that candidate.

Concretely, the aggregate is the sum of each criterion's score times its weight, divided by the sum of the weights. So a candidate who scores well on the heavily weighted primary criteria ranks above one who only does well on secondary ones, even if their raw scores look similar.

Because the aggregate is a weighted average on a 0–5 scale, the candidate list ranks by genuine fit, not by how many profiles matched a keyword. Sort by score to bring the strongest matches to the top, then read the per-criterion justifications before shortlisting.

Editing scores

You can override an individual criterion score for a candidate — for example, if you have context the profile doesn't capture. When you edit a score, Recruiter recomputes that candidate's aggregate using the same weighting, so the candidate re-ranks immediately. You can also add a note alongside the edit.

What triggers re-scoring

Scoring is idempotent — it always works on candidates that don't yet have an aggregate score, so re-running it is safe. It re-runs automatically when the candidate pool or the scorecard changes:

Change	Effect on scoring
Edit the query or JD	Filters re-extract, candidates re-fetch, then everything re-scores.
Edit filters	Candidates re-fetch and re-score.
Edit scorecard criteria or types	Existing candidates re-score against the new scorecard; no re-fetch.

You can also re-score a single candidate against the current scorecard on demand from their profile. Re-scoring runs in the background; the list polls and updates, so give it a moment after a change before the new numbers appear.

What's next

Candidate Search — how candidates get into the search in the first place.
Outreach — reach out to the candidates that scored well and you've shortlisted.