Ask the desk
Grounded analytics from the on-device model (Qwen on the DGX Spark — nothing leaves the LAN). It answers only from the desk's own data feed, quotes numbers verbatim with their caveats, and says so when something isn't in the data. Name metros for study-level detail.
Try asking
Each chip is grounded in the current feed: house views, Card Counter, trade-out validation, buy-vs-build quadrants, and flip timing.
Generated answers — verify anything decision-critical against the cited pages and studies. The model reads the desk's research; it does not add outside facts. ~10 tokens/sec on-device: answers stream in over ~30–90 seconds.