Recruiting Scorecards: Study Version Drift Before Comparing Interview Results

Research question

On August 18, 2026, what must a recruiting record show before an HR team compares interview feedback across candidates or roles? This study treats each submitted scorecard as a versioned process artifact. A rating is not comparable merely because two forms use the same number. Questions, anchors, stages, interviewer roles, and completion dates can differ. The research distinguishes evidence that a process was followed from the hiring manager’s judgment about a candidate.

Methodology and evidence scope

The method compares EEOC recordkeeping requirements, NIST Privacy Framework principles, GAO internal-control concepts, and the Uniform Guidelines on Employee Selection Procedures. It maps those public sources to scorecard versioning, access, retention, and review. The sources support provenance and controlled evidence; they do not prove that a score is fair, job-related, predictive, or lawful. This is general operational information, not selection-law advice or a substitute for qualified counsel.

The comparability problem

A requisition can remain open while a question is clarified, a rating scale is renamed, or an interviewer adopts a private document. A late score may therefore represent a different instrument than an early score. Missing comments, edits after submission, and interviews conducted at different stages add further ambiguity. A neat average can conceal that the inputs were produced under unlike instructions. Administrative research should expose those differences before anyone interprets the numbers.

Evidence design

A useful record links requisition, candidate token, interview stage, scorecard version, question set, rating definitions, interviewer role, completion time, missing-field state, and edit history. Maintain a version register with owner, approval date, effective stage, and retirement date. Preserve corrections as events rather than silently replacing the original. A coordinator can verify that a form is complete and linked to an approved version; the recruiter and hiring manager decide what the feedback means.

What the evidence supports

Version control supports a narrower claim than “the best candidate scored highest.” It can show whether feedback was requested, which instrument was used, whether a required field was missing, and whether the process changed midstream. It can also reveal calibration meetings, reviewer disagreement, and late edits as process events. Those signals may justify a qualified review, but they do not establish bias or candidate quality. The honest output is a bounded comparison with caveats.

Measures and safeguards

Measure the share of submissions tied to an approved version, missing-field rate by stage, elapsed time between interview and submission, late-edit count, and the number of requisitions with midstream version changes. Report counts with the defined denominator. Limit candidate data to authorized users and the hiring purpose. Avoid copying sensitive notes into an administrative dashboard. A clean completion rate cannot demonstrate fairness, so process measures must not be marketed as outcome evidence.

Role boundary

Outsourced HR support may maintain the version register, send an approved reminder, check completion, preserve the submitted record, and flag drift. It should not rewrite interviewer comments, rank candidates, infer protected characteristics, choose a rating, advise on accommodation, or declare a selection process compliant. The hiring manager, recruiter, HR, and legal owner retain selection authority. Escalate any request to alter evidence or bypass an approved scorecard.

Limitations and conclusion

Role requirements, accommodations, local law, interviewer training, and record systems differ. Public guidance cannot determine whether a particular scorecard is valid or whether a decision was discriminatory. Versioned records nevertheless make comparisons more honest: they show whether the underlying inputs were actually comparable. The evidence-led conclusion is to study version drift before comparing interview results and to keep administrative completeness separate from selection judgment.

Route-specific analysis

For an outsourced HR team, the practical comparison begins before anyone summarizes scores. First identify the requisition and stage, then verify that the scorecard version was approved for that stage. Next compare rating anchors and questions rather than only the numeric result. A missing field can be an administrative exception, while an unusual rating is an interviewer judgment that should not be rewritten. If a form was edited after submission, retain the event and ask the authorized recruiting owner whether the record remains usable. If the scorecard changed during the requisition, report the change and avoid a single blended average unless the owner has a documented reason. This treatment protects candidates and hiring teams from false precision. It also gives the client a clear handoff: the support record says which evidence exists, where it differs, and what requires judgment. It does not say which candidate should advance. That boundary is especially important when a requester asks an administrator to clean up comments or normalize ratings. Cleaning structure is permitted; changing a person’s substantive assessment is not. A version register therefore becomes a research instrument, not merely an archive.

Sources

EEOC recordkeeping: https://www.eeoc.gov/employers/recordkeeping-requirements. Uniform Guidelines: https://www.uniformguidelines.com/. NIST Privacy Framework: https://www.nist.gov/privacy-framework. GAO Green Book: https://www.gao.gov/green-book. These sources frame provenance and limitations; they do not rank candidates.

Sources

1. EEOC recordkeeping requirements
2. Uniform Guidelines
3. NIST Privacy Framework
4. GAO Green Book

Related Research

Benefits Eligibility: Reconcile Events Before Asking a Plan Owner to Decide

Employee-Relations Intake: Measure Routing Without Repeating Sensitive Narratives

HRIS Integrations: Investigate Failed Handoffs Before Replaying Records