An Observation Is Evidence; an Action Rationale Is a Decision Record

A trading trace says, “Volume rose, so we must buy.” The first clause describes a possible observation. The second asserts an obligation. A useful record shows how the evidence became a decision and which mandate, if any, supplied that obligation.

We separate observation from action rationale because each needs a different evaluation. A market fact needs source support and time. A trade needs an authorized objective, sizing logic and policy checks. Combining them into one persuasive paragraph can hide the missing step.

Read four sentences separately

Consider an illustrative trace: “Hourly volume increased 30%. Price remains below resistance. The mandate allows a breakout entry after confirmation. We will wait for a confirmed break.” The first two sentences refer to data. The third refers to authority. The fourth is the selected decision under those inputs.

A reviewer can now ask focused questions. Was the volume comparison calculated over comparable intervals? Was resistance derived from a stated method? Did the active mandate require confirmation? Did the action preserve that condition? One failed answer need not invalidate every other component of the trace.

The percentage is illustrative and supplies no market forecast. Its function is to expose the data dependency. A faithful explanation should retain the actual interval and timestamp so a later replay can reconstruct what “increased” meant.

Earlier rationale has a different status

Our published controls record documents pre-launch traces invoking invented rules. A combined intervention removed law-like wording, treated prior decisions as context, and prohibited fabricated named rules and thresholds. The affected sell-trace measure fell from 57% to 3%, with exact comparison-arm counts absent from the published account.

That result motivates a narrow interpretation: earlier reasoning needs explicit historical status. A previous “wait for confirmation” choice can remain informative without becoming a universal requirement for every future strategy. The compound intervention prevents attributing the entire change to one memory label.

Our state and memory article expands the provenance requirement. Stored observations and past decisions should carry timestamps and source identities so the runtime can assemble current evidence without silently renewing old authority.

Score support and compliance apart

For a proposed evaluation, split each rationale into sourced facts, model interpretations, mandate references and selected action. Check factual support against the saved snapshot. Check mandate references against the compiled version. Then validate the typed action independently.

A persuasive rationale can accompany an invalid order. A terse rationale can accompany a policy-valid order. This separation helps diagnose whether the problem lies in explanation quality, evidence use or execution authority. It also limits conclusions about model reasoning, since visible text is an observation of output rather than a complete measurement of the model's internal process.

DXAP describes reasoning behind recorded decisions. We would evaluate that surface by how well readers can connect evidence to an authorized action, including waiting.

At the frontier of delegated trading, a rationale earns its value by making a decision testable. It should tell us which facts mattered, which conditions controlled the choice and what would need to change before another action becomes appropriate.