ARCHITECTURE
Evidence-first pipeline
User question
Ask about an asset, a move, and a horizon (today / 24h / week).
AI planner (OpenRouter)
Generates search angles and coverage constraints (recency, macro, catalysts).
Bright Data SERP
Gets fresh sources across news and web with consistent parsing.
Web Unlocker
Fetches the pages that matter (and that usually block bots).
AI extraction + linking
Summaries, entities/actors, catalysts, edges, spillover hypotheses.
Supabase
Stores sessions, pipeline events, and evidence so the UI can replay the trace.
Dashboard
Breaking Tape, Sources, Narratives, Evidence Map (Graph/Mind/Flow), Price, Video Pulse, Chat.
Bright Data is used as the evidence acquisition layer; the model uses that evidence to reason and produce artifacts.
Architecture Notes
Production intent for the current product shell
DATA LAYER
Bright Data provides SERP and page extraction; AI reasoning uses only collected evidence.
ARTIFACT LAYER
The run produces evidence, tape, graph links, and narrative clusters, then persists replay snapshots.
UI LAYER
Map-first workspace with timeline and inspector, plus snapshot replay from dashboard history.