01 / PROPOSE
Research produces a candidate action.
The research provider is replaceable. The candidate arrives with evidence, risk boundaries, and an explicit requested action.
AI can recommend the action without owning the authority.
TraderLab uses a paper-only financial workflow because the stakes are easy to understand. The real product is the control pattern: evaluate the request, enforce policy, require approval when needed, and leave a receipt.
live risk: blockedbroker action: not submittedpolicy: reviewmode: paper
02 / GATE
Policy determines what may advance.
GREEN, YELLOW, ORANGE, and RED states separate low-risk review from human approval, trigger confirmation, or a hard block.
03 / RECEIPT
Every meaningful decision leaves evidence.
Allow, deny, approval requirement, execution boundary, and review state become inspectable records instead of disappearing into an agent log.
What the public can see
The control contract.
- → AI recommendations do not equal permission.
- → Sensitive actions can require explicit human approval.
- → Live execution remains blocked in this proof.
- → The same pattern applies to email, files, support, network changes, and account operations.
What stays private
The operator plane.
- → Candidate symbols, scores, and strategy details.
- → Gate violations and approval requirements.
- → Receipt identifiers and review records.
- → Protected settings and control-request generation.
- → Broker credentials and production authority are never exposed.
Visible safety loop
proposal → evidence check → strategy gate → human approval when required → scoped action → receipt → review
The authenticated dashboard lives on intignai.dev. Public visitors receive the product story and proof boundary—not private operational detail.