Job Bro adds pessimistic job scoring

This is a useful correction to the usual resume-matcher problem: false confidence is worse than a low score when candidates are deciding where to spend attention.

• –Domain caps are the right primitive because seniority does not transfer cleanly across embedded, infra, ML platform, fintech, and web roles
• –Stealth and seed-stage risk detection makes the product more practical than pure skill matching, especially for candidates with compensation floors
• –The sub-50ms target is ambitious if the system is doing real contextual reasoning, so the interesting technical question is how much logic is deterministic versus LLM-driven
• –The biggest v0.1.9 opportunity is explainability: candidates need to see which resume evidence did or did not satisfy the domain gate