Technical reference for the current detection buckets, evidence model, and local-first extension behavior.
Distinguishes workflow-like interfaces and helps separate AI tools from AI agents when the evidence supports it. Known platform matches still help, but unknown pages can also be recognized from workflow behavior instead of hostname lists alone.
Scans for scripts, widgets, metadata, and content clues that suggest AI involvement. Strong claims are reserved for stronger structural evidence so the extension does not overstate what it sees.
Checks AI-related image hints and provenance signals where available. Image findings are intentionally conservative so the popup can signal useful evidence without claiming proof that is not there.
Groups privacy findings using OBSERVED, DISCLOSED, and POSSIBLE so direct technical signals, site-stated language, and softer inference stay clearly separated in the UI.