The practical answer
Extract the role’s actual tasks and required evidence, avoiding irrelevant personal attributes. Compare a résumé claim with its supporting example rather than adding unsupported achievements. Use the same interview criteria, record evidence for each score and leave uncertain items unresolved for review. Before assigning a rating, write the observed task, the candidate’s own contribution and the result they can substantiate. Distinguish a quoted answer from the interviewer’s interpretation so a later reviewer can trace the rating to evidence. Before comparing candidates, check that the rubric version and question scope are consistent. If evidence is insufficient, record the missing information and the follow-up needed instead of inventing a precise score or averaging incompatible assessments.
Plan the input and the review
- Extract the role’s actual tasks and required evidence, avoiding irrelevant personal attributes.
- Compare a résumé claim with its supporting example rather than adding unsupported achievements.
- Use the same interview criteria, record evidence for each score and leave uncertain items unresolved for review.
- Before assigning a rating, write the observed task, the candidate’s own contribution and the result they can substantiate. Distinguish a quoted answer from the interviewer’s interpretation so a later reviewer can trace the rating to evidence.
- Before comparing candidates, check that the rubric version and question scope are consistent. If evidence is insufficient, record the missing information and the follow-up needed instead of inventing a precise score or averaging incompatible assessments.
Choose the operation for this step
- Inspect résumé relevance: Compare highlighted keywords and missing sections with the actual résumé; treat the result as a checklist, not an automated hiring decision or prediction.
- Record interview evidence: Define scoring anchors before interviews, record examples behind ratings and distinguish absent evidence from a demonstrated lack of skill.
An editorial example to compare
Make the acceptance condition visible
A missing keyword becomes a zero score, even though the résumé describes the same task in different words.
The reviewer checks the described task, records the matching evidence and uses a consistent follow-up question for unresolved experience.
What 33% keyword coverage establishes
Use an illustrative résumé and specify three custom keywords: SQL, Python and stakeholder communication. The résumé says only, “Maintained a MySQL reporting database; presented monthly results to sales managers.” The current check searches normalized résumé text for normalized keyword substrings. MySQL contains SQL, so one of three matches and coverage rounds to 33%. Python does not match, and a description of a communication task is not automatically a literal match for stakeholder communication.
Coverage can therefore miss equivalent wording and overstate what a substring establishes. Return to the complete account: what database maintenance involved, what the candidate personally owned, and who used the monthly reports. Someone with relevant experience can describe the task clearly; adding unperformed skills or invented metrics merely to make every keyword match would destroy the evidence.
Make a problem-solving rating observable
For a role involving attendee-list cleanup, define an identical invented exercise in advance: two people share a name but have different IDs, one ID repeats, and one date is missing. Assess whether the candidate preserves both people, identifies the repeated ID and explains the missing date, rather than whether they name a particular tool brand. The example below anchors 1, 3 and 5; define intermediate performance for 2 and 4 before interviewing.
An evidence note should say “flagged the repeated ID and retained both rows pending business confirmation,” rather than “seemed careful.” If the missing-date question was not asked, mark that evidence unresolved and use a consistent follow-up. An unasked question is not observed failure.
| Anchor | Observable performance |
|---|---|
| 1 / 5 | Deletes a same-name person without checking IDs or explaining lost records |
| 3 / 5 | Identifies the repeated ID and preserves sources; missing-value treatment needs a follow-up |
| 5 / 5 | Distinguishes shared names from duplicate keys, explains missing values and reconciles output rows |
Review every default 3 before saving a comparison
New scorecard criteria currently default to 3/5. The displayed mean includes every criterion; an empty evidence field does not exclude a rating. In an illustrative card with 5, 4 and an untouched 3, the mean is (5 + 4 + 3) / 3 = 4.00. An unassessed third criterion does not establish mid-level performance.
There is no separate unassessed/excluded rating state here. Record what remains unassessed and the follow-up in the evidence field, treat the card as a draft, and complete the evidence before comparison. If a criterion should not be assessed, revise the same rubric version for all candidates instead of removing one person’s weak item to raise their average. Read exported averages together with criteria and evidence rather than using them as automatic decisions.
Check the final result and preserve evidence
The EEOC explains protected characteristics in US employment. Use job-related evidence and check the rules that apply to your own context.
Check a generated summary against its source before relying on it. Preserve original wording while cleaning application records.
Acceptance checks
- Each criterion relates to the role and was defined before evaluating candidates.
- Scores retain supporting notes and uncertainty.
- The exported comparison receives a human review before a consequential decision.
Common questions
Does a high match score predict job performance?
This check does not establish predictive validity. Review actual experience and job-related evidence.
Can missing evidence automatically count as failure?
Distinguish an unanswered question from observed performance and apply the same follow-up process consistently.
Should a résumé be rewritten with invented metrics?
No. Use only achievements and measurements the person can support.
Further reading & sources
Put the guide into practice
Open the relevant tool, preview the result and check your exported file.
