From question to saved report

Six tools, one study

The assistant translates your intent into tool calls. You should still review the choices and evidence. You do not need to memorize JSON parameters to use the guided conversation.

1. Register: connect the dataset to your question

register_dataset

Use a private upload or a shared sample. Give the assistant the file identifier, not a path on your laptop.

2. Inspect: understand what the data can support

inspect_dataset

Review units, periods, outcome, treatment timing and missing information. Confirm the variable roles.

3. Plan: make the choices before the fit

plan_analysis

Specify the effect you want, comparisons, assumptions, uncertainty and analysis purpose. Review the plan.

4. Run: execute the approved analysis

run_analysis

Confirm the exact plan. The server runs the analysis in your workspace and records a durable run identifier.

5. Interpret: read the evidence in context

get_result

Explain the size of the effect, how precisely it is estimated, which comparisons support it, and what the diagnostic checks found.

6. Report: keep a record you can revisit

render_report

Generate a report from saved results. Reading or rendering a completed result does not fit the model again.

The three research checkpoints

1. Research specification: question, causal structure, variable roles, target, assumptions and specification.
2. Plan and execution: review the actual immutable plan and give permission to compute it.
3. Evidence and claim: review what the saved result supports and the exact wording of the proposed conclusion.

Your assistant will ask for the required approvals at each checkpoint. You can revise the question or plan before committing to the computation.

Exploratory and confirmatory work

Choose quick to explore a question with a lighter analysis. Choose confirmatory when the study needs the full supported inference, comparison and sensitivity requirements. Make that choice before fitting so the plan includes the right work.

The planner explains whether those requirements can be met by your design. After the run, the report shows which checks were completed and any unresolved findings. A change of purpose requires a new plan.

How long will it take?

Runtime depends on sample size, method, resampling and sensitivity work. The current server admits one heavy fit at a time; other approved analyses can remain queued while analysts continue planning and reading results.

Record the run identifier. Check the existing run instead of repeatedly submitting a new one. A reconnect or page refresh is not a reason to create a duplicate study.

Keep the assistant on the research path