For the person at the window

The demo window is an agent over the Limma MCP server with a page in front of it. Written so far:

• Search mode — what the window's search mode does: the agent sends a brief to the search verb first and reads the closed set from its warrant, and how the result set is laid out.
• People search — a brief about people, a niche included: what the agent does with it, how to read the result card and the side-by-side with a vector search, a window in front of a directory of people, and how a run is scored against the demo corpus's answer key.
• The Q&A view — the page a customer is sent: one question box, the answer drawn from what the search returned, a follow-up box. What it says and never says, the routes it leaves open, what a follow-up is handed, the example questions and how to read them ahead.

How to read a result set (the count, the receipt chip, the None and Not recorded lines, the provenance split), what "which pages" means, the tested questions per workspace and the do-not-ask list are recorded per workspace in demo/RUNBOOK.md, which remains the operating record for the window; the concepts behind each line are in the integrator pages (a count, None versus Not recorded, search).