evidence-sourcinglisted
Install: claude install-skill tberghane/calibratr-plugins
# Evidence-First Candidate Sourcing
## Overview
Source candidates across many platforms at once, and return **evidence, not assertions**. Every
candidate ships with verbatim profile URLs and concrete proof-of-work artifacts (each with its own
URL and a quantified signal). A name with no links and no work samples is not a sourced candidate.
Core principle: **show the work**. Recruiters trust a repo, a paper, or a launch they can click far
more than a one-line claim.
## When to Use
- "Source 20 backend engineers who've built distributed systems"
- "Find ML researchers working on RAG / retrieval"
- "Who has shipped developer tools / launched on Product Hunt in dev infra?"
- Building a pipeline for a role, or expanding a thin pipeline with new sources
**Not for:** scoring/ranking against a rubric (use `proof-of-work-scoring`), deep single-candidate
verification (use `candidate-dossier`), or turning a JD into search criteria (use `role-dna` first).
## The Iron Rule: Verbatim URLs Only
**NEVER construct, infer, or guess a profile URL — especially LinkedIn.**
- Only include a URL if a tool returned it **verbatim** in its raw result.
- A name-based slug like `linkedin.com/in/jane-smith` is a guess. Guesses resolve to the wrong
person or a 404 and destroy recruiter trust. A missing link is honest; a wrong link is not.
- If no verified URL exists for a field, leave it out and write **"URL not verified"** in output.
- This applies to you AND to every tool result you summarize