A Number in a Trading Mandate Needs a Unit and a Trigger

“Keep risk low” leaves several decisions open. An agent can interpret it as small size, fewer positions, short holding periods, or tight stops. Replacing the phrase with a number helps only when the number has a defined denominator and a condition that determines when it applies.

We study this distinction because an executable mandate must survive more than a language-model reading. It also has to supply values that policy can check. Our mandate compilation article connects strategy language to typed controls and source attribution.

Rewrite one sentence carefully

An illustrative instruction begins: “Risk 2% on good opportunities.” Assume account equity is $20,000. Two percent equals $400, but the sentence leaves the target quantity unresolved. It could mean $400 of position value, $400 of margin, or a $400 estimated loss if a stop executes at its reference price.

A clearer proposal is: “For a new position, planned loss at the specified stop must be at most $400, using current reconciled account equity as the 2% denominator. Recalculate before submission.” This remains an example rather than a complete strategy. Gaps, fees and slippage can produce realized loss beyond the planned amount, so the control name should say precisely what it estimates.

The source map would preserve equity timestamp, percentage, reference stop, quantity formula and rounding direction. If one input is missing, an explicit unresolved result gives the owner a concrete question to answer. Guessing the denominator would make a confident-looking instruction less reliable.

What the published cohort tells us

Our DX Terminal Pro paper companion reports that users with concrete exit conditions or parameter changes achieved profitability 4.2 times as often as users asking the agent to outperform or pick winners. This was an observational comparison in one bounded 21-day market. Owner engagement and strategy choice can influence both instruction specificity and outcomes.

That result motivates a controlled question: does adding units and conditions improve mandate fidelity on the same saved scenarios? It supplies a reason to test specificity, while causal investment claims require a stronger design.

For an initial replay, compare a vague instruction with the explicit version using identical account states and model settings. Count unresolved interpretations, limit violations and appropriate requests for clarification. Track action frequency separately. A system that takes fewer trades after receiving a clearer risk constraint may be interpreting the mandate more faithfully.

Numbers can become quotas

Specificity also introduces a failure mode. A sentence saying “about three trades when opportunities justify it” can become an obligation to trade three times. The compiler must distinguish a descriptive expectation from an authorized minimum or maximum. Units alone cannot supply that distinction.

DXAP describes strategy refinement through chat. That interaction is a useful place to expose the resolved meaning of a number before it becomes active. The public product description does not establish the result of this proposed specificity test.

We would judge the rewrite by whether the agent and validator apply the intended constraint under difficult scenarios, including a turn with no qualifying opportunity. Better prose is the input. Faithful behavior is the measured outcome, and trading profitability remains a separate question.