Developer tool

Retrieval validation

Simulate how a retrieval-augmented agent ingests this site. Enter a query to see which knowledge-pack chunks match, which concepts are hit, and a confidence score. Runs entirely in your browser against the published export pack.

Try: accountability who owns guardrails autonomy level evidence threshold AI-SaaS responsibility

Raw endpoint response (JSON) 
Run a query to see the JSON response.

About this tool

The corpus is built from /export/framework.json (definitions plus concept labels). Scoring uses BM25 over tokenized chunks. A low confidence score or empty concept hits for a term that should exist signals a chunking gap or a missing semantic link worth fixing at the source. Append &format=json to any query URL to get the JSON response only.