No Trade Is a Decision Worth Recording

A trading turn ends without an order. That can mean the agent found no qualifying opportunity, lacked current information, or failed before deciding. Counting all three as successful restraint would conceal operational problems and make an inactive agent look unusually disciplined.

We treat no-trade as an explicit decision when the model selects it under a valid mandate and sufficient state. Our mandate compiler framework includes an abstention path beside transaction actions, so waiting has a defined place in the action contract.

Three reasons, three records

In an illustrative turn, the mandate requires a breakout plus confirming volume. Price approaches resistance while volume remains ordinary. A no-trade decision can record that the condition is unmet and identify what would change the answer. This is a completed market decision.

In a second turn, the portfolio snapshot is stale. The runtime withholds new exposure pending reconciliation. That is a state-quality response with an explicit reason and a recovery path. It may be appropriate behavior, but it says little about the model's ability to identify opportunities.

In a third turn, a research tool times out and the runtime never produces a typed action. This is incomplete work. Its visible absence of orders resembles the first turn, while the causal path and next action differ.

Keeping these reasons distinct makes the record useful for owners and researchers. “Wait for volume confirmation,” “refresh account state” and “retry failed research” explain different kinds of progress without burying the user in internal detail.

An opportunity test needs an opportunity

We would evaluate abstention with paired saved scenarios. One has the required confirmation and acceptable costs. The other changes only the confirmation field so the entry condition is unmet. Under a mandate authorizing that entry, the expected behavioral contrast is a permitted proposal in the first and a reasoned wait in the second.

A model that waits on both cases may have low violation rates and poor mandate responsiveness. A model that trades both may be ignoring the condition. Scoring both halves prevents universal inactivity from earning credit as selective judgment.

The DXRG benchmark card article describes why the evaluation subject and population must be stated. Action rate without opportunity conditions cannot distinguish a selective policy from a broken runtime.

Historical controls and current product scope

Our operating-layer controls paper companion reports historical action-frequency gradients across user settings. Those gradients describe behavior within a bounded deployment. They leave the economic value of a particular abstention unresolved, because avoided losses and missed gains require a stated comparison.

DXAP explicitly describes recording no-trade turns. That product choice gives waiting an observable place in the loop. It does not, by itself, establish that abstention improves returns.

We would retain the state identity, active condition and selected reason for each completed wait. An owner can then see why the agent held still, while an evaluator can reproduce the decision. The next harness improvement can target missing data, failed tools or decision quality separately rather than optimizing a raw trade count.