Build and run a study
A study asks one question of a definition's events and answers it across every one of them: How often did the break reach the prior day's high, and does it depend on how far above VWAP it started? You ask in plain words, the agent writes the study, and nothing runs until you approve it.
A study measures events the definition has already found, so run the definition over history first. The Agent offers to when they are missing.
Ask the Agent
Open Agent at the top of the sidebar.
Ask the question, naming the definition and what you want measured: Using my BTC opening range definition, how often did breaks reach the prior day's high before the day ended? Split it by distance from VWAP.
Approve the Plan proposal card. If the definition has no events yet, a Data collection proposal card comes first: click Collect.
When the study is written, a Study run proposal card asks Run the study on BTC? and shows the Instrument, Timeframe, Range, Events in scope and Profile. Check the event count is what you expect, then click Run.
The chat shows the run's progress. When it finishes, a Study results card appears. Click Open the full report to see it beside the chat.
Approving a run covers its scope. If the study hits a bug, the agent fixes the code and runs it again over the same events without asking a second time.
Test your first trading idea follows this route from the first message to a Forward Insight.
Build it on the definition
Open the definition from Tools → Definitions and open the Studies tab.
Click New study. The page asks What do you want to study?
Type the question and press Enter, or start from one of the Suggested studies.
The agent writes the study and shows each change as a Study edit card. Click Show diff to see the code or Preview to try it, then Accept. To turn it down, click Reject, say what should change and click Send reason.
When the agent proposes a run, pick the datasets under Datasets, check the Run window and any Parameters, and click Run. With several datasets the button counts them, such as Run (3). You can pick up to 8.
The run card moves through Queued, Loading events, Running and Finalising, then shows the results. With more than one dataset, each gets its own tab.
Choose how much you want to see
The coaching mode under the message box sets how the agent works with you. It is saved for each study.
Mode | How the agent works |
|---|---|
Guided | Talks about your question and the results, and keeps the code out of the way. The default |
Collaborative | Designs the measurements and charts with you |
Technical | Precise about the study's design. Code stays in the preview card |
To see or edit the script yourself, click Show Code. Code, Lints and Manifest show the study, Save draft saves your edits, and Run Study asks the agent for a run proposal to approve. Scripting covers how studies are written.
Asking a good question
Name the outcome and the time limit: reached the prior day's high before the session close.
Say how to split it: by direction, by distance from VWAP, by day of week.
Ask for what happened first when both matter: the 1R stop or the 2R target.
When it goes wrong
Attach a definition to this study chat before running: the study was started without a definition. Start it from the definition's Studies tab.
The dataset picker says No datasets on this definition yet: run the definition over history first.
The card says Run failed: ask the agent in the chat what went wrong.
Studies are kept on the definition's Studies tab, for the profile you have selected. Switch to All studies to see every profile's. How many studies you can keep depends on your plan: 25 on Starter, 1,000 on Pro and unlimited on Ultra. How many runs can go at once is on Runs at once.