What Research and Chart Subagents Add to an Agentic Trading Turn

A trading decision often contains several smaller questions. What happened? How did price respond? Does the proposed position fit the user's mandate? Giving one model every question at once can make the reasoning hard to inspect. DXAP describes research and chart subagents. We judge that architecture by the work each specialist returns.

Give a worker a question it can finish

Consider an illustrative event-driven thesis. A research worker checks the event's original source and identifies its timing. A chart worker examines the asset's recent price structure. The final decision-maker combines those results with the portfolio and active constraints.

Each worker should have a bounded task. Asking a research worker to 'find reasons to buy' selects evidence in advance. Asking it to establish the announcement's content, timestamp, and unresolved conditions produces a result that can support buying, waiting, or rejecting the thesis.

The chart worker likewise needs a defined observation window and instrument. Otherwise it can describe whichever timeframe best supports a conclusion. Its output should identify the examined data and the conditions under which its interpretation would change.

Agreement can share one cause

Two specialist outputs agreeing feels persuasive. Their agreement may still derive from the same headline, the same market snapshot, or the same model family. We would preserve evidence lineage so the final agent can recognize those shared causes.

For example, the research worker describes a bullish announcement, while the chart worker interprets the announcement-driven price spike as momentum. The second output offers a market reaction, rather than independent evidence about the announcement's eventual effect. That distinction changes how confidently a combined thesis should be expressed.

Our continuous-record research is relevant background for evaluating decision quality. Adding workers introduces an architectural hypothesis: decomposing the task may improve completeness or accuracy. The hypothesis deserves a comparison against a simpler turn using the same information.

Test synthesis when workers disagree

A useful fixture gives the research worker a favorable event and the chart worker a deteriorating price structure. The final agent should explain how the disagreement affects its decision. Another fixture makes the research source stale while current price appears attractive. The final explanation should preserve the source-age issue rather than smoothing it into a confident story.

We would compare evidence coverage, contradiction handling, and turn cost. Decision scoring comes afterward. Better research organization can help owners understand a strategy even where its economic benefit remains unsettled.

The controls paper companion explains the surrounding enforcement architecture. Specialist analysis remains upstream of execution authority. The final submitted action should still pass the same policy checks regardless of how many workers contributed.

What the user should see

A compact explanation can show each specialist's finding and the reason the final agent accepted or discounted it. The full worker output can remain available for inspection without filling the main account screen.

That is the platform advantage we would pursue: deeper analysis with a readable decision surface. The architecture earns its place when workers resolve concrete uncertainties and the final action remains accountable to the user's strategy.