ProTrader

A model handed a raw trade history will pattern-match and start sounding confident about the future. Handed a computed set of metrics such as win rate, average loss, and drawdown, it has something concrete to talk about and far less room to speculate.

The coach's job is to explain a member's own behaviour back to them, which is genuinely useful and does not require a prediction. What it must never do is tell someone what to buy.

Explicit guardrails. The coach is constrained away from trade recommendations and toward reviewing performance, because a platform that lets an AI hand out entry signals takes on a liability nobody wants.

This is a product decision with real teeth: it shapes the prompt, the tool surface and what the interface will render.

A daily cap per member. Without one, a single enthusiastic user can generate an unbounded bill, and coaching is a feature people will use heavily on a bad trading day.

Responses stream, which also matters commercially. Streaming makes a long answer feel immediate instead of appearing to hang, so members do not retry and double the cost.

Broker accounts feed an eight-tier ladder, so rank reflects verified account activity rather than self-reported claims. Compliance rules sit on the same data, which keeps progression honest.

That verification is what makes a community ladder mean anything. A leaderboard anyone can type their way onto is decoration.

A control plane provisions each white-label tenant automatically: its own dedicated database and its own domain, with no manual setup step.

Giving each tenant a separate database rather than a shared one with a tenant column costs more per tenant, and buys hard isolation. For a platform holding trading data, that trade is worth making.