Source-grounded reasoning artifacts

Evidence,
with edges.

Doubt keeps support, contradiction, qualification, and missing evidence inspectable. Every evidence card opens the exact dated source region behind it.

Try the playground Explore Skills vs MCP Star on GitHub ↗

Current map Use Agent Skills for reusable procedure and context. Add MCP for live tools and resources. Combine them when the workflow needs both.

2public maps

10source regions

12/12contract cases

0runtime deps

Inspect the reasoning

Four ways in.

Validate a map without installing anything, inspect a real AI architecture question, audit the product decision, or test maps against cited prose.

Zero-install

Validate a map locally

Paste or drop JSON, inspect fail-closed findings, render the same interactive artifact, and download portable JSON or HTML.

No upload · no account · no telemetry ↗ Topical map

Agent Skills vs MCP

Which boundary fits reusable procedure, which fits live capability, and why “both” is sometimes the only honest answer.

5 sources · 2 contradictions · 1 unknown ↗ Dogfood map

What should Doubt become?

The evidence that killed a launch-video direction, preserved its strongest advantage, and exposed the missing reader test.

Open the decision record ↗ Open benchmark

Does the map beat the memo?

Five paired topics, the same frozen sources, local-only assignment, and no result claim before ten complete readers.

Take the reader study ↗

supports

contradicts

qualifies

missing