admin/intelligence
Evidence, graph and learning read models; curated relationships and observed behavior remain distinguishable.
Files and their individual responsibilities are listed in the generated source inventory. Each file starts with its contract summary.
Modules
ExternalKnowledge.tsx: Private connected-app evidence browser shows active-source provenance without exposing it to shoppers.KnowledgeExplorer.tsx: Product-centred evidence inspector reads canonical facts and graph relationships beyond overview sampling.KnowledgeFacts.tsx: Canonical catalogue facts shown alongside graph evidence; prices and inventory come from current product state.KnowledgeOverview.tsx: Whole-shop knowledge census, operational next steps and provenance activity; examples never masquerade as learned facts.KnowledgePreview.tsx: No-model evidence preview makes customer/private retrieval boundaries and missing facts inspectable.KnowledgeSources.tsx: Searchable cursor source library, guarded lifecycle decisions and single-language source editing.KnowledgeView.tsx: Unified knowledge workspace connects sources, product evidence, observations and no-model retrieval previews.MemoryView.tsx: Durable co-purchase evidence and revision-bound merchant decisions, with simulation labels and explicit confirmation.SourceEditor.tsx: Single-language source editor with inherited fields, product lookup and private-first API/file ingestion.knowledge-types.ts: Typed knowledge read models preserve source ownership, revisions, sampling and privacy boundaries.
Verification
Run npm run build, npm test, npm run test:coverage and npm run architecture from frontend/. Coverage includes untested source files; a module in this folder is not automatically fully tested. See the root testing guide for backend integration and coverage limits.
The connected cognition editors use native APIs: EvidenceReview.tsx reviews
current document-bound claims; GuardrailSettings.tsx edits canonical commerce
settings and currency scales; ExperimentStudio.tsx registers and controls native
layout experiments with clear preliminary/final readouts. Each uses the shared
typed four-language vocabulary. Revision and permission failures remain visible.
See cognitive commerce for tested scope.