What problem does it solve? Research reports often mix measured facts with guesses that look identical on the page, so unverified numbers get quoted as truth downstream. This Skill forces every finding to carry a provenance grade — MEASURED, READ, INFERRED, or UNVERIFIED — and its renderer refuses to produce a report from a badly graded record. ## Core Features & Use Cases - Graded findings: Every claim is tagged with one of four closed-set grades, with MEASURED and READ requiring an explicit Evidence line citing the command, conditions, or source. - Refusing renderer: A Python script validates the record and generates a self-contained HTML view with inline SVG charts (provenance mix, bars, line, matrix, range), writing only outside the repository. - Use Case: Investigating whether a CI gate is slow, you record each finding with its grade, render the report, and share a page where the reader sees at a glance that three findings were measured and one was never checked. ## Quick Start Ask the agent to research a question using the evidence-first-research skill and produce a dated, graded record with an HTML view.
