resume-screening-rubric

---
name: resume-screening-rubric
description: "Design structured, fair CV/resume screening rubrics and score resumes against them with consistent, explainable verdicts. Use when the user asks to screen, score, rank, shortlist, or evaluate resumes/CVs against a role, or to build a screening scorecard for recruiters or an ATS."
---

# Resume Screening Rubric

## Purpose
Replace gut-feel CV review with a structured rubric — the same discipline AI-ATS projects automate: parse → extract → score → rank → explain.

## Rubric design (before scoring anything)
Score each candidate 0–4 on 5–7 criteria derived ONLY from the JD's must-haves:
| Criterion | Weight | 0 | 2 | 4 anchors |
|---|---|---|---|---|
| Core skill match | 30% | none | partial, adjacent | direct, recent, at required depth |
| Relevant impact | 25% | duties only | some metrics | quantified outcomes at similar scale |
| Domain/industry fit | 15% | none | adjacent | direct |
| Seniority calibration | 15% | 2+ levels off | 1 level off | exact |
| Trajectory & stability | 10% | red flags | mixed | consistent growth |
| Education/certs (only if required) | 5% | missing | partial | met |

## Scoring rules
- Evidence-based only: if the CV doesn't show it, it's a 0 — never infer from company brand or school prestige.
- Knock-out checks first (work authorization, mandatory license/cert) — binary, then rubric.
- Two independent screeners on a sample; recalibrate if ratings diverge >1 point.
- Decision bands: ≥80% shortlist, 60–79% hold/talent pool, <60% decline (with respectful template).
- One-line justification per criterion — explainable screening is mandatory for AI-assisted tools (EU AI Act) and good practice everywhere.

## Bias guards
- Hide/ignore: name, photo, age, address, marital status when scoring (anonymized screening where possible).
- Penalizing employment gaps is disallowed unless the criterion is skill recency — then test skill directly.
- CV language quality is NOT a proxy for ability unless the role requires it.

## Outputs
- Rubric table (markdown/CSV)
- Per-candidate scorecard: criterion scores, weighted total, band, justifications, interview-probe suggestions
- Batch mode: ranking table of all candidates + shortlist recommendation