Describing a permissions model does not convince anyone. Being made to type a confirmation before anything is permitted to act is a different experience.
EMIL Lab is the real control surface, running against simulated data. You set the analysis permissions, the markets, the risk limits, the maximum exposure, the automation level and the trading permissions. Then you request arm, read what you are authorising, and type it.
What we deliberately made awkward
- Arming requires typing. There is no one-click arm, because arming is an act of authorisation.
- The authorisation text is specific. It states what will be permitted, the hard limits, and that the intelligence layer cannot remove market risk.
- The dialog does not close by clicking outside it. Dismissing an authorisation prompt by accident should not be possible.
What we deliberately made easy
- Disarming. One button, always rendered, high contrast, no confirmation. The Escape key works too. Withdrawing authorisation should never require navigating anywhere.
The log narrates what the layer is doing in readable language — including refusals, which are the interesting part. An order rejected for breaching a concentration limit tells you more about a system than one that filled.