Scheduled Trading Reviews and Agent-Triggered Revisits Solve Different Problems

A strategy can need both regular attention and a response to an unusual event. A schedule answers when to review again. A trigger answers which changing condition deserves an earlier review. DXAP describes both scheduled turns and triggers. We think the distinction becomes clear when a position is already open.

A review time is different from an event

Imagine an illustrative position with a scheduled hourly review and a condition to revisit after a material price move. The hourly review provides coverage even when the market stays quiet. The event condition allows the agent to examine new information sooner.

Neither clock alone determines the right trade. The scheduled turn may conclude that the original thesis remains valid. A triggered turn may reach the same conclusion after inspecting the move. Waking the agent is a request for a decision with current state, rather than an automatic instruction to transact.

This distinction keeps the user's strategy readable. The user can specify why a condition matters without treating every crossing as an order. It also lets an agent request a future check while keeping the reason attached to the request.

Two wakeups, one account

Suppose the price condition fires just before the scheduled review. Both wakeups refer to the same account, possibly the same market observation. A runtime needs to coordinate their access to current positions and unresolved orders.

Our preferred fixture makes the first turn submit a proposal while the second wakeup arrives. The test checks whether the later turn uses reconciled state and recognizes any unresolved action. Simply counting both turns as successful would miss a duplicated position change.

Another fixture moves price through the trigger repeatedly. The test asks whether the trigger has a defined lifecycle and whether repeated observations create repeated requests. A strategy can intentionally allow repeated reviews, but that choice should be explicit and measurable.

An agent's follow-up needs a reason

An agent might decide that an event requires confirmation later. We would preserve the condition it intends to check, the observation that motivated it, and any expiry. This creates a compact research question for the next turn.

A stale trigger illustrates why that record matters. If the owner changes the strategy before the condition fires, the runtime should evaluate the wakeup against the current mandate. The historical follow-up request can remain in the trace without taking precedence over the owner's newer instruction.

Our controls research provides the authority background, while the continuous-record paper connects repeated turns to historical fleet observation. Here the narrower contribution is a lifecycle test for two kinds of attention.

Measure attention as well as activity

We would report scheduled reviews, event reviews, duplicate wakeups, and resulting actions separately. A rising turn count could reflect useful responsiveness or redundant work. Only the joined record explains which.

A platform with both clocks gives owners more ways to express how a strategy should stay attentive. The benefit is greatest when each wakeup carries its cause and each decision starts from a consistent account state. That is the coordination problem frontier agentic trading has to solve.