Atomic requirements
One observable requirement per entry, linked to the wording that supports it.
A visible chain from role brief to recruiter judgment.
Role Evidence keeps the requirement, resume citation, review gap, and final override in one inspectable record for staffing and HR technology workflows.
Evidence, not a score
0External enrichment sources
0Uploaded documents retained
HumanFinal employment decision
The requirement ledger comes first.
A job description is separated into atomic requirements before any resume is reviewed. Each entry keeps the source wording, essential or preferred classification, review status, and requirement-set version.
The author confirms the classification and the job-related rationale. Role Evidence does not invent a rationale or turn ambiguous wording into a firm requirement.
Author review
Unclear classifications stay visible until a recruiter confirms or edits them.
Versioned set
A changed requirement creates a new reviewable version instead of rewriting history.
Evidence binding keeps the source attached.
A supported statement points to exact text in the candidate-supplied resume. The server verifies that each citation range exists and that the excerpt matches the redacted source before a result can say evidenced or partially evidenced.
When no verified citation exists, the result is not evidenced or needs review. It does not claim that the candidate failed the requirement.
Evidenced
The submitted resume contains a verified citation that directly supports the requirement.
Partially evidenced
A verified citation supports part of the requirement, with the remaining detail called out.
Needs review
Parsing, classification, or citation uncertainty is routed to a person.
The submission dossier is built for client review.
The dossier groups each requirement with verified evidence, unresolved details, and the recruiter decision. It gives the client a review path without rewriting the candidate resume or hiding uncertainty.
Every evidence dossier carries the statement: Evidence output requires human review.
The decision record preserves context.
Versions
Requirement-set, model, model revision, prompt, and output-schema identifiers.
Bindings
Evidence verdicts, verified citations, explanations, and extraction reliability.
Human action
Reviewer, recruiter override, override reason, decision stage, and timestamp.
Lineage
Suppression manifest and previous-record hash for a reviewable history.
Blindness controls reduce irrelevant signals.
Before inference, the live analysis workflow suppresses names, contact details, street and postal addresses, gendered pronouns, graduation years, school names where degree context can remain, photographs, and citizenship or national-origin markers.
Only category names and counts return in the suppression manifest. Suppressed values are not returned, retained, or used for cross-customer learning.
The human-review boundary is a product rule.
Role Evidence does not calculate an overall candidate score, rank people, predict success, infer personality or culture fit, or make an employment decision. Evidence output cannot be the sole basis for an employment action.
Inspect the workflow with your own authorized files.
The live analysis workflow keeps the result in browser memory and requires a verified email before one analysis.