Resume Screening Rubric: 12 Criteria with Human Review

Quick answer

Use the 12 criteria only after a qualified team documents essential functions, validates each criterion as job-related and necessary, defines equivalent evidence and accommodations, removes protected or proxy features, anchors ratings, trains at least two reviewers, and establishes notices, audits, records, correction, and appeal required for the location and tool. AI may retrieve and propose evidence; it should not silently invent facts, convert missing evidence into failure, or make the final employment decision.

Before scoring: establish the selection contract

Job analysis and essential functions

Name the actual tasks, conditions, outputs, frequency, consequences, and minimum level. Separate essential from marginal functions and preference from necessity; record who approved the analysis and when it expires.

Decision states, not one opaque score

Use demonstrated, partially demonstrated, insufficient evidence, not assessed, accommodation pending, contradicted, and not applicable. Missing or inaccessible evidence is not automatically a zero.

Human authority and independence

Reviewers must see source evidence, disagree, correct, request accommodation, and stop the process. The final decision owner records reasons independently instead of rubber-stamping a rank or recommendation.

Twelve job-related criteria with review controls

Not every role should use all 12. Activate only criteria supported by the job analysis; publish the same requirements and evidence routes to applicants; version any change before screening begins.

Essential-function evidence

Question: Which documented essential function does this evidence support, at what required level, and by what job analysis?

Acceptable evidence: Resume statements, portfolio or work-sample evidence, verified role scope, and structured follow-up tied to the same published requirement.
Do not infer or proxy: Do not infer ability from title, employer prestige, disability disclosure, assistive technology, appearance, or a missing conventional career path.

Mandatory license or credential

Question: Is a current credential legally or operationally required on the start date, and is an equivalent or obtain-by date allowed?

Acceptable evidence: Credential type, jurisdiction, issuer, status, expiry, verification source, equivalency rule, and documented start-date necessity.
Do not infer or proxy: Do not convert a preference into a knockout, assume foreign credentials are invalid, or request unrelated medical, age, citizenship, or protected information.

Job-relevant knowledge

Question: What knowledge is required to perform the function, and can the applicant demonstrate it through more than one route?

Acceptable evidence: Specific concepts, tools or regulations named in the approved role profile, plus equivalent education, employment, volunteer, military, self-directed, or portfolio evidence.
Do not infer or proxy: Do not use school rank, degree brand, jargon density, native-language style, or keyword count as a proxy for knowledge.

Relevant task experience

Question: Has the applicant performed comparable tasks at the required complexity, recency, and ownership—not merely held a matching title?

Acceptable evidence: Task, context, constraints, decisions, collaborators, duration, scale, outcome evidence, and what the applicant personally owned.
Do not infer or proxy: Do not penalize career gaps, part-time work, caregiving, contract work, unemployment, or shorter tenure without a validated job-related reason.

Work-product quality

Question: Does submitted work demonstrate the published quality standard under comparable conditions?

Acceptable evidence: Authorized portfolio artifact or standardized work sample scored with anchored criteria for correctness, completeness, safety, usability, and reasoning.
Do not infer or proxy: Do not require unpaid production work, expose confidential prior-employer material, compare unequal accommodations, or score visual polish when irrelevant.

Problem framing and judgment

Question: Can the applicant identify the task, constraints, unknowns, tradeoffs, escalation point, and verification method?

Acceptable evidence: Structured example or work sample showing assumptions, alternatives, risk, evidence, decision, and what would trigger review or rollback.
Do not infer or proxy: Do not infer judgment from confidence, charisma, fast response, accent, personality labels, or agreement with the reviewer.

Communication for the actual role

Question: Can the applicant produce and understand the communication the essential functions require, using an accessible format or reasonable accommodation where applicable?

Acceptable evidence: Role-specific writing, explanation, listening, documentation, language requirement, audience adaptation, and confirmation behavior assessed consistently.
Do not infer or proxy: Do not score accent, grammar variation, speech pattern, camera behavior, facial expression, disability-related communication, or native language unless demonstrably job-related and lawful.

Collaboration and handoff evidence

Question: Has the applicant coordinated dependencies, resolved disagreement, shared context, and accepted accountability in work similar to the role?

Acceptable evidence: A structured example identifying participants, authority, handoff package, conflict, applicant action, decision, feedback, and result.
Do not infer or proxy: Do not reward similarity to the current team, social style, extracurricular access, referral status, or vague “culture fit.”

Customer or stakeholder responsibility

Question: Can the applicant identify the affected stakeholder, preserve rights and commitments, communicate uncertainty, and close the loop?

Acceptable evidence: Role-relevant case with stakeholder need, authority, evidence, risk, decision, communication, acceptance criteria, correction, and follow-up.
Do not infer or proxy: Do not favor customer-facing polish over accuracy, pressure applicants to disclose protected experiences, or treat emotional style as empathy.

Safety, compliance, or control discipline

Question: Where the role truly requires it, can the applicant follow controls, recognize limits, document evidence, escalate, and recover from error?

Acceptable evidence: Job-specific policy or scenario, required control, authority boundary, incident response, audit record, correction, and learning evidence.
Do not infer or proxy: Do not use generic “risk aversion,” criminal-history assumptions, credit information, medical data, or unrelated background signals as a shortcut.

Learning and adaptation evidence

Question: Has the applicant learned a relevant capability, incorporated feedback, or adapted responsibly when requirements changed?

Acceptable evidence: Initial gap, learning plan, resources, feedback, revised behavior or artifact, verification, and limits that remained.
Do not infer or proxy: Do not equate youth, recent graduation, frequent job changes, hobby technology use, or fast tool adoption with learning capacity.

Documented work constraints and availability

Question: Can the applicant meet the essential schedule, location, travel, authorization, or physical requirements as lawfully and narrowly defined, with accommodation considered where applicable?

Acceptable evidence: Published requirement, business reason, start date, permitted eligibility evidence, applicant-confirmed availability, accommodation route, and exception authority.
Do not infer or proxy: Do not infer family status, religion, pregnancy, disability, age, nationality, commute reliability, health, or willingness from address, name, gap, photo, or social profile.

Use anchored ratings that preserve uncertainty

Demonstrated

Source evidence directly meets the published anchor at the required level; reviewer cites the exact span or artifact and no material contradiction remains.

Partially demonstrated

Relevant evidence exists but scope, recency, complexity, ownership, or verification falls short of the anchor; record the precise gap and a consistent follow-up question.

Insufficient evidence or not assessed

The record does not yet support a rating, or the criterion was not evaluated. This is not failure; use the same opportunity to provide evidence where the criterion is necessary.

Accommodation pending, contradicted, or not applicable

Pause when an accommodation may change how evidence should be gathered. Keep conflicting evidence visible for review. Remove criteria that the approved job analysis does not support.

Worked example: a career gap is not evidence of weak reliability

This hypothetical example is not a real candidate or validated hiring result. A resume shows a two-year gap, contract operations work, and a portfolio describing an incident handoff. An unsafe screen assigns a reliability penalty for the gap and a prestige bonus for employer names. The governed rubric removes both proxies.

Map evidence to the function. The reviewer links the incident artifact to collaboration, stakeholder responsibility, and control discipline—not to a general personality score.
Preserve uncertainty. Ownership and scale are unclear, so those criteria are partially demonstrated with structured follow-up rather than silently filled by AI.
Do not investigate the gap. The gap is ignored because it is not a validated requirement. Reviewers do not seek medical, family, disability, age, or other protected explanations.
Record a reviewable decision. Two reviewers cite the same anchors, reconcile disagreement, document evidence and remaining gaps, and route the next authorized assessment. No AI rank becomes the decision.

How OpenMax can coordinate human-reviewed screening

OpenMax can coordinate approved job versions, applicant notices, permitted resume intake, redaction where appropriate, accommodation pauses, evidence retrieval, rubric proposals with exact citations, independent reviewer assignments, disagreements, corrections, audit samples, deadlines, and authorized next steps. Humans retain job analysis, legal scope, criterion validation, accommodation decisions, sensitive evidence, thresholds, interviews, and final employment decisions.

1 · GovernRole, law, purpose, owner, notices, audit

2 · ConfigureCriteria, anchors, equivalents, accommodation

3 · ReviewSource evidence, uncertainty, two reviewers

4 · DecideHuman reason, authority, correction, challenge

5 · MonitorSelection rates, errors, accommodations, drift

Legal, fairness, privacy, and accessibility boundaries

Do not use the template until qualified counsel determines applicable employment, discrimination, disability, privacy, labor, recordkeeping, automated-decision, notice, audit, and candidate-rights duties for every location and tool.
Do not collect or infer protected traits or proxies; keep voluntary demographic monitoring data separated from decision makers where required; secure resumes, model prompts, reviewer notes, accommodations, and audit records.
Provide an accessible process and a clear accommodation route; requesting an accommodation must not lower a score. Test each tool for screening out qualified people with different disabilities and provide an alternative where required.
Do not treat a bias audit, impact ratio, four-fifths heuristic, vendor assurance, or human click as proof of legality or fairness. Validate job relevance, inspect the full selection process, investigate differences and intersectional/low-volume limits, preserve challenges, and stop harmful use.

Sources, editorial method, and limitations

OpenMax editors reviewed current U.S. EEOC materials on job-related selection and prohibited practices, U.S. DOJ guidance on disability discrimination and accommodations in algorithmic hiring, New York City DCWP requirements for covered automated employment decision tools, and NIST AI RMF Core outcomes for governance and human oversight. We synthesized the 12-criterion rubric, state model, and hypothetical career-gap example. Sources were rechecked September 3, 2026.

Scope note This global template cites selected U.S. examples and is not a complete jurisdictional survey. Guidance may not be binding and requirements can change. Sources do not validate the rubric, define every role, certify any tool, or guarantee lawful or fair outcomes. Test the actual role, process, population, tool, accommodations, and decisions under current applicable law.

Frequently asked questions

Can AI automatically reject a resume?

Do not assume so. Determine applicable law and tool scope, validate the procedure, provide required notices and accommodations, audit impacts, preserve meaningful human authority, and offer correction or challenge.

Should career gaps lower the score?

Not unless a qualified process establishes a narrow, validated job-related requirement—which is uncommon. Evaluate current evidence for essential functions instead of seeking protected explanations.

Can the same rubric be reused for every role?

No. Criteria and anchors must follow the approved job analysis for the specific role and context. Version and revalidate changes.

Does human review make an automated tool safe?

Not by itself. Review must be informed, independent, empowered, timely, documented, and able to change or stop the outcome; the full process still needs validation, accessibility, impact monitoring, and legal compliance.

What can OpenMax automate?

It can coordinate approved criteria, evidence retrieval, exact citations, uncertainty, reviewer assignments, accommodation pauses, corrections, audit records, and authorized next steps while humans retain validation and employment decisions.