The caller's OWN concept/algorithm map — the Workspace canvas.

Built from ONLY this caller's data, through the SAME two key-bound stores the rest of the client plane reads: org_library.list_for (their saved algorithms + their org's company-shared ones) and dispatch_runs.list_for (their own run traces). No owner-plane source is touched — /owner/map is the internal concept map and must never reach a customer.

Envelope mirrors the cockpit's flat typed graph so one renderer serves both: {caller, nodes:[{id,label,kind,parent,scope,runs,steps,ts}], edges:[{from,to,type,meta_type}], counts:{...}} kind: "algorithm" (a saved artifact) | an imported Library kind | "step" (a step inside one) | "run". meta_type: "abstraction" (contains) | "causal" (step order) | "associative" (a run that matches a saved algorithm's title). Key-bound, allowlisted, empty-safe — an empty workspace returns empty lists, never a stub.