Evidence-Based Optimization
How Pathos turns model output into a source-verified, evidence-grounded resume instead of trusting a rewrite because it sounds convincing.
A resume optimizer can raise a keyword score by copying the posting, inventing a metric, changing a title, flattening a career into one target role, or quietly deleting material that does not help the match. Those are easy numerical wins and serious product failures.
Pathos therefore does not define success as better prose or a higher score. The output has to improve alignment while preserving the candidate's identity, chronology, evidence, section coverage, and ability to defend every claim in an interview.
Every full optimization begins with a source contract built from the candidate profile and the sections the user chose to include. The prompt may receive a relevance-selected view of that material, but the contract retains the role identities and source content the response must be reconciled against.
User-approved skills and explicit evidence supplements are separate authorization inputs. Outcome correlations and old cross-user patterns receive no document credit. A useful signal about what tends to work is not evidence that this candidate personally did something.
What the system guarantees
- Generated content is treated as an untrusted proposal and verified against a versioned source contract before it can become the working resume.
- Identity, dates, credentials, metrics, skills, bullets, section density, and summary claims have separate enforcement paths instead of one generic hallucination check.
- The artifact is checked and rescored after downstream cleanup and keyword recovery, so the visible score describes the resume that actually ships.