Which Interview Feedback Record Should a Hiring Team Trust?

Research question: which record a coordinator should use when interview notes, scorecards, chat messages, and hiring-system fields disagree during a Philippines hiring process? This protocol examines a narrow administrative evidence problem. It does not make an employment, legal, payroll, security, hiring, benefits, or performance decision.

Why this matters to a staffing buyer: A buyer may receive a complete-looking candidate summary even though one interviewer used an obsolete rubric, another edited a score after discussion, and a third left feedback in a private message. The operational question is not who should be hired. It is whether the hiring owner can see the origin, timing, version, and limits of the evidence used for that decision.

Evidence basis: the Philippine Data Privacy Act and its implementing rules establish principles including transparency, legitimate purpose, proportionality, accountability, security, access, and retention. National Privacy Commission materials describe the responsibilities surrounding personal-data processing. DOLE materials provide labor-context references. NIST CSF 2.0 and SP 800-53 provide governance, identity, access, audit, personnel-security, and record-integrity concepts. These sources frame a test; the responsible organization must determine its actual duties.

Unit of analysis: one interviewer submission for one candidate at one defined interview stage. Fixing the unit before review prevents an attractive batch total from concealing one unresolved person, permission, record, or decision. Every unit receives a stable non-identifying reference.

Method: construct ninety synthetic submissions spanning completed scorecards, missing criteria, late edits, duplicate interviews, withdrawn comments, conflicting recommendations, inaccessible attachments, wrong-candidate notes, panel consensus recorded without individual inputs, and free text containing unnecessary personal data. Use invented people, organizations, dates, values, and identifiers only. A study administrator retains the seeded answer key separately until reviewers finish their first classifications.

Required evidence fields: candidate reference, requisition, stage, interviewer role, rubric version, question or criterion, original response, submission timestamp, later edit, edit reason, source link, access class, conflict code, and hiring-owner disposition. Each field must have a stated business purpose and authoritative source. Blank, unknown, not applicable, restricted, and not yet received remain different values; none may be translated into approval or completion for convenience.

Before testing, the responsible owner writes the accepted states, required evidence, role permissions, response windows, escalation path, and stop conditions. Freeze those definitions for the first pass. If a rule changes, version it and rerun affected cases instead of rewriting earlier results.

Primary measure: accurate preservation of the source, version, and sequence of each submission without converting interviewer observations into a coordinator recommendation. Secondary measures include mismatched candidates, hidden edits, missing feedback, unsupported consensus, excess disclosure, and unresolved conflicts. Reviewers cite the exact event supporting every classification and use cannot determine when evidence is insufficient. A confident guess counts as an error even when it accidentally matches the answer key.

Error taxonomy: A numerical score does not prove that the required evidence was supplied. A later edit is not automatically more authoritative than a timely original. A panel summary cannot stand in for individual records unless the hiring owner explicitly defines that rule. Administrative completeness and candidate suitability are different judgments.

Decision boundary: The coordinator may distribute the approved rubric, monitor receipt, check observable completeness, preserve source links, and route conflicts. Interviewers own their observations. The hiring manager or authorized hiring panel owns weighting, interpretation, selection, rejection, accommodation decisions, and the final employment decision.

Comparison design: Compare a provenance-preserving register with a single overwritten summary field. Also compare coded missing-feedback reminders with circulation of candidate narratives in broad chat channels. Give reviewers the same underlying cases in randomized order. Compare correctness, unnecessary access, unresolved work, serious errors, and review time rather than measuring speed alone.

Privacy and security treatment: Use fictional candidates and minimum necessary attributes. Keep demographic, health, family, compensation, and background-check material outside the coordination view unless a specifically authorized process requires it. Inspect exports, notifications, meeting notes, and deleted-record recovery as well as the main interface. Record who viewed, exported, notified, or changed evidence because a restricted main screen does not prevent a downstream copy from exposing the same data.

Negative controls are essential. Include ordinary cases that should proceed, incomplete cases that should stop, and misleading cases containing a plausible but insufficient signal. A process that never stops is uncontrolled; a process that stops every case is not useful.

Analysis plan: Report source-matching accuracy, version accuracy, and conflict-routing accuracy by interview stage. Treat wrong-candidate association, unauthorized disclosure, and invented consensus as serious failures even when the final status happens to match the answer key. Two reviewers independently classify an overlapping sample. Preserve disagreements, categorize their causes, and send them to the named owner. Do not average disagreement away or let the first data entry become the answer by default.

Set quality thresholds before opening the answer key. The owner defines acceptable routing accuracy, maximum unresolved age, serious-error classes, and pilot stop conditions. A faster workflow fails if it increases unauthorized decisions, disclosure, false closure, or loss of provenance. Publish counts and denominators for every result, including exclusions and their reasons.

Run a repeatability check after the first review. A second reviewer receives the written definitions and clean cases, without coaching from the first reviewer. Low agreement may show an ambiguous rule, weak source, or missing state. Clarify the rule, record a new version, and retest the affected cases before live use.

Provider evidence should match the proposed operating model. Buyers can request a sanitized role demonstration, blank register, permission view, escalation map, and sample audit export. Each artifact has a date and scope. A policy, sales statement, or successful demonstration does not establish continuous operation; missing evidence stays an open question.

Separate fact, classification, decision, and inference in the result. A source event is an observed fact. Applying a frozen rule produces a classification. An authorized owner may make a decision. Researchers may then infer where the workflow is fragile, but that inference must be labeled and cannot be presented as a measured live-company outcome.

Uncertainty should remain visible. Missing events, ambiguous definitions, unavailable owners, integration delays, inaccessible evidence, and contradictory sources receive explicit codes. Report waiting time separately from active processing time so an external dependency is not mistaken for coordinator performance.

Limitations: Synthetic interviews cannot reproduce interpersonal bias, lawful accommodation duties, every recruiting platform, local retention requirements, or the quality of a real interviewer’s judgment. The protocol tests record provenance and routing, not whether a candidate should advance. A live pilot should begin with a small approved queue, named reviewers, least-privilege access, monitored exceptions, and a stop rule for unexpected sensitive information or decisions outside the written lane.

Decision use: The decision-grade output is a feedback provenance register showing each source, rubric version, timing, later change, unresolved conflict, and hiring-owner disposition while leaving the selection decision outside the coordinator lane. Buyers can ask a provider to demonstrate the record with sanitized cases. That demonstration is point-in-time evidence, not a guarantee of compliance, accuracy, security, or employment outcomes.

Sources checked September 23, 2026: National Privacy Commission, Implementing Rules and Regulations of the Data Privacy Act of 2012 (https://privacy.gov.ph/implementing-rules-regulations-data-privacy-act-2012/); National Privacy Commission, Republic Act 10173 – Data Privacy Act of 2012 (https://privacy.gov.ph/data-privacy-act/); Department of Labor and Employment Bureau of Working Conditions, Labor Advisories (https://bwc.dole.gov.ph/issuances/labor-advisories/); National Institute of Standards and Technology, Cybersecurity Framework 2.0 (https://www.nist.gov/cyberframework); National Institute of Standards and Technology, Security and Privacy Controls for Information Systems and Organizations, SP 800-53 Revision 5 (https://csrc.nist.gov/pubs/sp/800/53/r5/upd1/final). These primary government sources provide privacy, labor-context, governance, access-control, and audit concepts. They do not decide a specific employment matter, certify a provider, or replace advice from authorized legal, HR, payroll, security, or benefits owners.