cv-and-humanlisted

# CV and Human Polish a CV so it clears automated screening while staying truthful. The strategy is grounded in the **source code** of a real LLM screener (read directly, not from commentary): ATS scoring splits into a **knowable, controllable surface** — parseability, literal keyword coverage, project links, real open-source framing, the explicit bonus triggers — and a **noisy LLM-judgment layer** you can't pin down (whether your projects are "worth" 18 vs 24 points swings run to run). Lock down the controllable surface; feed the noisy layer the strongest *true* material; never promise a score. Note that some screeners take **no job description at all** and grade against a fixed rubric — so this skill handles both JD-matching and JD-less tools. See `references/ats-mechanics.md`. ## Core principles (read before editing anything) 1. **Never fabricate.** Do not invent skills, titles, dates, employers, metrics, or open-source work. If a JD requirement is genuinely absent from the candidate's background, surface it as a gap for them to address — do not paper over it. Tailoring means truthful reframing and surfacing, not lying. This is both an integrity line and a practical one: fabrications get caught at interview and burn the candidate. 2. **Win the deterministic surface first.** Exact skill/keyword coverage and clean parseability are checklist logic — they score consistently. This is where effort reliably moves the needle. 3. **Strengthen, don't promise,