Verify the citation before trusting the verdict.

The job description is normalized, then separated into atomic requirements. Essential or preferred classifications are used only when the source wording supports them. Otherwise the requirement remains unclear and needs review.

Every rationale status requires recruiter confirmation. The system does not invent a business justification for a requirement.

The server removes specified names, contact details, addresses, photographs, gendered pronouns, graduation years, school names where degree context can remain, and citizenship or national-origin markers before inference.

The result includes only the category names and suppressed counts. It never returns the removed values.

System instructions and uploaded document text stay in separate fields. Any instruction found inside the job description or resume is treated as document content and cannot change the analysis rules.

When the submitted resume does not contain a verified citation, the result is not evidenced or needs review. That statement is limited to the submitted file. It is never a conclusion that the person failed the requirement.

Document parsing can fail, role wording can be ambiguous, suppression can remove useful context, and model output can be incomplete. The live analysis workflow performs no OCR and uses no external enrichment. Its output cannot be the sole basis for an employment decision.