build-evidence-map

Build validated JSON evidence maps linking sources, claims, and counterevidence to a decision.

38.5k|4.9k|Updated Jun 11, 2025
One-click install
npx skills add https://github.com/github/awesome-copilot --skill build-evidence-map
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: build-evidence-map
Source: https://github.com/github/awesome-copilot/tree/main/skills/build-evidence-map
Command: npx skills add https://github.com/github/awesome-copilot --skill build-evidence-map

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

Contested technical decisions often collapse supporting and contradicting evidence into unstructured prose, losing track of what was actually sourced, what disagrees, and what remains unknown. This Skill turns one contested question into an auditable, machine-validated evidence map.

Core Features & Use Cases

  • Structured reasoning graph: Models a decision as typed nodes (position, claim, evidence, unknown) connected by typed edges (supports, contradicts, qualifies, missing) with plain-language notes.
  • Bounded source regions: Every evidence node records URL or local path, publisher, dates, a page/section/line/timestamp locator, and a 40-500 character checkable excerpt.
  • Fail-closed validation: A bundled Node.js validator checks schema invariants, rejects cycles, duplicate edges, unused sources, and confidence percentages, then issues a SHA-256 receipt.
  • Use Case: When reviewing a proposal to migrate databases, map the benchmark results, vendor claims, and missing production-load data into a .doubt.json artifact so reviewers can see exactly what supports the verdict and what could overturn it.

Quick Start

Ask Copilot to build an evidence map for your contested technical decision, citing exact source regions, and validate the resulting .doubt.json file with the bundled script.

Frequently Asked Questions about build-evidence-map

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I create an evidence map for a contested technical decision?

Frame one falsifiable question and a provisional position, then collect bounded source regions with URLs, dates, locators, and excerpts. Atomize reasoning into position, claim, evidence, and unknown nodes connected by supports, contradicts, qualifies, or missing edges, and write the result as a .doubt.json file.

How do I validate a .doubt.json evidence map?

Run node scripts/validate.mjs decision.doubt.json with Node.js 18 or newer. The validator uses only Node.js built-ins, requires no npm install or network access, and prints VALID followed by a 64-character receipt when the map passes all checks.

When should I use an evidence map instead of a fact-checking workflow?

Use an evidence map when relationships between evidence, intermediate claims, trade-offs, and missing facts matter for a consequential decision. For simple factual claims or general fact-checking, a verification workflow such as doublecheck is the better fit.

What source information does each evidence node require?

Each source needs a title, publisher, publication date, retrieval date, URL or local path, a bounded locator such as a page, section, line range, or timestamp, and a checkable excerpt of 40 to 500 characters. Sources without an exact locatable region become unknown nodes instead.

Why does evidence map validation reject confidence percentages?

The schema forbids confidence fields to avoid false precision in reasoning. Uncertainty is expressed structurally instead by adding an unknown node, narrowing the position, or attaching a qualifies edge to a claim.

Can I render or verify an evidence map against live sources?

Rendering to HTML requires the user to have already installed [email protected], and source verification with doubt verify retrieves recorded HTTP sources only with explicit network permission. Neither command is run implicitly, and verification results are never written by hand.