Risk controls

A safety filter that is switched on and cannot be evaluated blocks the trade. Missing evidence is never read as approval.

Fail closed, always

This is the single most consequential decision in the risk model. When a data source is unavailable, an automated system has two options: proceed as though the check passed, or refuse. Proceeding is how automated traders buy things nobody would have bought — the filters appear to be on, the interface shows them enabled, and they are silently doing nothing. Every enabled filter here refuses on missing data.

Enforced together, not one at a time

Limits are checked at the instant capital is committed, so several calls arriving in the same second cannot each be told the balance is free and collectively exceed a ceiling none of them individually breached. A bot set to three open positions opens three, not thirty, however many calls land at once.

Stopping

Every bot has an immediate stop that halts new entries while continuing to manage exits on positions already open. There is also a platform-wide pause for genuine incidents. Neither abandons money already in the market — stopping the strategy and abandoning the position are different actions, and a system that conflates them is dangerous in exactly the moment you need it most.