Ask your agent what would invalidate its thesis

Ask for evidence that could invalidate the thesis before asking the agent to defend it more eloquently. The useful answer identifies an observable contradiction and explains its consequence under the strategy. We would review that answer against the recorded evidence, then decide whether a strategy change is warranted.

The chat guide describes conversations grounded in the selected agent's strategy, account, and recent activity. Chat itself places no trade or triggers no trading turn. This distinction matters when an owner wants criticism of a thesis without authorizing new execution.

A fictional debrief after a long entry

Illustrative dialogue: the owner asks, “You described demand acceleration. Which source establishes that, and what observation would show the acceleration stopped?” The agent names a metric and says a decline in the next completed measurement would weaken the claim.

The owner follows with, “Was that condition part of the saved strategy at entry, or are we defining it now?” This separates a previously active criterion from a useful new proposal. A retrospective improvement belongs to the later review, with the earlier turn preserved as recorded.

The third question is, “If the measurement becomes unavailable, what does the current mandate require?” A strategy can require waiting for dependent new entries while treating existing exposure separately. The owner needs that distinction spelled out instead of accepting “remain cautious” as a complete rule.

Make the falsifier inspectable

A good falsifier specifies the instrument, measurement, timeframe, and consequence. “If momentum weakens” is a topic. “If the next completed window fails the defined threshold, stop considering additions and review the existing position” is a candidate criterion whose later use can be checked.

That example supplies strategy language for review rather than an instruction already active on an account. Configured policy fields still constrain admissible orders. Our mandate compiler article explains the relationship between prose and controls.

Avoid rewarding the sheer number of objections. Three weak hypothetical objections can be less useful than one decisive contradiction supported by an original source. Ask the agent to rank which observation would most change its decision and why.

Review the later decision without outcome hindsight

Save the criterion before the next eligible observation. Then compare the later recorded turn with what became available at that time. Did the evidence actually cross the defined condition? Did the agent acknowledge it? Did the proposed action match the current mandate?

Our trace feedback article gives the framework for connecting those stages. Judge the falsifier against its predeclared evidence and the review against the information available. Profit and loss supply separate economic observations.

If the owner adopts a revised criterion, confirm it through the documented proposal flow and review the saved settings or active instruction. The result to seek is a thesis with an explicit failure condition and a record that makes its application observable.