The Evidence Extraction Method for Resume Scoring

Why Resume Scoring Fails Without Structured Evidence Extraction

Most resume reviews are impressionistic. A hiring manager skims a document, forms a gestalt judgment in under thirty seconds, and then works backward to justify it. The problem is not laziness — it is a structural one. Resumes are marketing documents, written to trigger positive affect, not to surface verifiable facts. Without a disciplined candidate evidence extraction procedure, reviewers are scoring the writing quality of the resume, not the capability of the person behind it.

Research on unstructured review supports this concern. Lievens & Highhouse (2003, Journal of Applied Psychology) found that aesthetic and surface features of application materials influence evaluator judgments independent of substantive content. More broadly, the case for structured, criteria-referenced evaluation over holistic impression has been made repeatedly in the selection literature — including in Schmidt & Hunter's (1998, Psychological Bulletin) landmark meta-analysis, which established that structured methods consistently outperform unstructured ones in predicting job performance.

This guide describes a specific, repeatable procedure: the Evidence Extraction Method. It is not a scoring rubric by itself — it is the data-collection step that makes any scoring rubric reliable. If you have already read [Candidate Evaluation Criteria: How to Score Candidates], think of this as the upstream process that feeds it.

Step 1: Lock the Evaluation Dimensions Before You Open the Resume

Action: Define, in writing, the specific dimensions you are evaluating before you read the first resume. Do not let the resume define the criteria.

Verdict organizes candidate evaluation across six dimensions:

Write these down before opening any resume. This prevents the common failure mode of discovering a dimension mid-review and retroactively applying it only to some candidates.

What good looks like: A one-page evaluation sheet with the six dimensions listed, each with a one-sentence definition tailored to the specific role. For a Head of Engineering role, "Capability" might be defined as: demonstrated architectural decision-making on systems serving >1M users.

Step 2: Read the Resume Once, Cold — Mark Only Claims, Not Judgments

Action: On first pass, highlight or annotate every claim the candidate makes. Do not evaluate yet. Use a simple tagging scheme:

• Q = Quantified claim (e.g., "grew revenue 40%")
• V = Verifiable claim (e.g., company name, title, dates, product launch)
• A = Attributed claim (credit claimed for a team result, unclear individual contribution)
• U = Unverifiable assertion (e.g., "strategic thinker," "strong communicator")

Why this matters: Mixing the reading step with the evaluation step introduces confirmation bias. Once you form an early impression, you unconsciously weight subsequent evidence to match it. This is a well-documented phenomenon — Nickerson (1998, Review of General Psychology) provides a thorough review of confirmation bias mechanisms in evaluation contexts.

What good looks like: A resume with margin annotations or a parallel document listing every distinct claim by type. A five-page resume might yield 40–60 tagged claims. U-tagged items should outnumber Q and V items significantly on most resumes — that ratio itself is informative.

Step 3: Map Each Claim to a Dimension

Action: Take your tagged claims and assign each one to the dimension it is most relevant to. Some claims will map to multiple dimensions — assign to the primary one and note the secondary.

This is where candidate evidence extraction becomes analytical rather than mechanical. A claim like "managed a team of 12 engineers across two time zones" maps primarily to Influence (scope of leadership) but secondarily to Capability (cross-functional coordination). A claim like "reduced infrastructure costs by $2.4M in Q3 2022" maps to Track Record (specific, quantified outcome).

What good looks like: A table or annotated list that shows, for each dimension, which claims are available as evidence — and which dimensions have no supporting claims at all. Absence is evidence too.

Step 4: Apply Evidence Quality Grades

Action: For each mapped claim, assign one of three evidence quality grades:

• Grade A — Specific and verifiable: Dated, quantified, attributable to an identifiable organization or product. Can be checked in a reference call or public record.
• Grade B — Plausible but unverified: Internally consistent with tenure and role, but lacks specific numbers or external anchors.
• Grade C — Assertion without support: Adjective-driven language, vague scope, or outcome claimed without mechanism.

Do not reward Grade C claims. They are indistinguishable between a strong and a weak candidate — every applicant can write "results-oriented" and "collaborative leader."

This step is the core of structured candidate evidence extraction. It forces the evaluator to ask: what would I need to confirm this claim, and does this resume provide it?

Step 5: Score Against Dimensions Using Only Graded Evidence

Action: Now — and only now — assign a dimension-level score, using only Grade A and Grade B evidence. Grade C claims do not contribute to the score.

A simple 1–4 scale works well:

The resulting six-dimension profile is your evidence-based resume score. It is not a hiring decision — it is a structured input to one.

What good looks like: A candidate with scores of [4, 4, 3, 2, 3, 2] across [Capability, Track Record, Trajectory, Influence, Domain edge, Risk surface] where lower Risk surface scores indicate fewer red flags. That profile tells a specific story that you can defend, revisit, and compare across candidates consistently.

Worked Example: Senior Product Manager Candidate

Raw resume claims (selected):

• "Led product strategy for a B2B SaaS platform" → Tagged: A (attributed)
• "Grew DAU from 12,000 to 47,000 over 18 months" → Tagged: Q + V
• "Collaborated cross-functionally with engineering and design" → Tagged: U
• "Launched three features that reduced churn by 18%" → Tagged: Q
• "Promoted twice in four years at Acme Corp" → Tagged: V
• "Passionate about user-centered design" → Tagged: U

Mapping and grading:

Scores: Capability: 2, Track Record: 4, Trajectory: 3, Influence: 1, Domain edge: 1, Risk surface: 3 (no concerning gaps; mid-tenure is reasonable).

Interpretation: Strong delivery evidence, weak influence and domain signal. Worth an interview with targeted questions on stakeholder management and product craft — not a pass, not a hire, but a specific diagnosis. For more on how to structure those questions, see [Forensic Interviewing: Structured Kit Generation].

Common Pitfalls to Avoid

1. Scoring the resume formatting, not the evidence. A well-formatted, visually polished resume can inflate impressions. The extraction method deliberately separates presentation from substance.

2. Treating quantified claims as automatically Grade A. Numbers can be fabricated or misleadingly scoped. "Increased revenue by 200%" from a base of $10K is not comparable to the same percentage from $10M. Grade A requires both a number and a verifiable context.

3. Penalizing candidates who are honest about scope. "Contributed to" is less impressive than "led," but it may be more accurate. Over time, rewarding inflation trains applicants to over-claim. The extraction method rewards specificity, which is the right incentive.

4. Applying the method inconsistently across candidates. If you extract evidence from five candidates but evaluate the sixth holistically because you liked their background, the method's value collapses. Consistency is the mechanism by which bias is reduced, not the rubric itself.

5. Conflating absence of evidence with evidence of absence. A resume that does not surface influence claims may be from a candidate who has real influence but did not know to write about it. Flag the gap; probe it in the interview. Do not automatically score it as disqualifying — calibrate the expectation to the role level.

For more on how these evidence gaps interact with AI-assisted screening, [AI Resume Screening vs Human Screening: The Evidence] covers where automated systems perform well and where human judgment remains essential.

Evaluate Candidates With Verdict

If this method resonates but the manual overhead feels unsustainable at volume, Verdict applies structured candidate evidence extraction against your specific job description — surfacing graded claims, flagging gaps across all six dimensions, and generating a defensible, evidence-cited comparison across your candidate pool. It is not a decision engine. It is a better instrument for the decision you still have to make. Run your next role through it and see what the evidence actually says.