Research can take its time; trade execution can't. The order that shouldn't go out may have been approved by exactly the human in charge. Who blocks it?
Orders, gating, block and release records
Timing state, market recalculation, incident post-mortems
Every block converts to a task automatically, with handling on record
A human approves the day's candidate buys per the pre-market plan.
Scheduled reminders drive @risk-gate to recompute market state mid-session; it detects an indicator flip.
The buys approved that day get blocked automatically with the reasoning stated; the market weakens that day, proving the block right — with zero human involvement throughout.
After one process incident where a verbal veto never reached the execution layer, the team stands up a three-layer CIO sign-off gate in minutes.
Three days after going live, the gate's first real action: it blocks by default a swap selling a heavily-in-profit position for a low-scoring new name; pulling the thread exposes two flaws in the trigger definition, and the human ultimately decides to shut the feature off entirely.
Market state recomputed on schedule during trading sessions; a flip triggers a block.
Every block converts to a task automatically: after the close, verify whether the block was right.
Periodically reconcile execution-layer configuration with the decision ledger, so verbal rules can't drift.