How Caliber Engine Reasons About a Trade
A walkthrough of the steps Caliber Engine takes between a market signal and a routed order, including where the model defers to your rules.
Caliber Engine is built on the idea that automation should be auditable. Every order it places can be traced back to a decision, and every decision can be traced back to inputs you control. Here is what happens between a market event and a filled order.
1. Context assembly
The engine starts by assembling context: current price action, your open positions, account-level exposure, time of day, and any rules you have configured for the symbol. This context is the working memory the model reasons over.
2. Candidate generation
From that context, the engine proposes a small set of candidate actions. Stay flat, scale in, scale out, hedge, or close. Each candidate is annotated with its expected impact on your risk envelope.
3. Constraint check
Before anything is routed, candidates are filtered through your hard rules: max position size, max daily loss, prop firm trailing drawdown, allowed hours. If a candidate violates a rule, it is dropped without ceremony. Rules win arguments with the model, not the other way around.
4. Selection and sizing
The engine selects the surviving candidate that best fits your stated objective for the account, then sizes it against current account equity rather than nominal targets. Sizing is the part most retail systems get wrong, so this is the step where the most engineering care goes.
5. Routing and reconciliation
The order is routed to your broker or prop firm via their official API. Fills are reconciled back into the engine's state so the next decision starts from ground truth, not a stale snapshot.
Why this design
The split between reasoning and rules is intentional. Reasoning lets the engine adapt to context. Rules guarantee that adaptation never crosses a line you set. Together they give you something a pure model or a pure rules engine cannot: an automated trader you can actually supervise.
Trading involves substantial risk of loss. Past performance is not indicative of future results. Caliber Engine automates execution against rules you configure; it does not eliminate market risk.