# Write from your data

Keep analysis inputs, scripts, tables, and manuscript results connected.

A results paragraph is easier to trust when the calculation behind it is nearby. Use a research project to keep the analysis procedure and its outputs alongside the manuscript.

## Set up the project

Choose the **Reproducible analysis** starter. [Link the data folder](/docs/linked-folders/) so the research tools can read the original measurements without moving them into the paper.

Record what the columns mean, their units, and any exclusions before asking for an analysis.

## Give the assistant a bounded job

Use `/oleafly-data-analysis` or an **Analysis** [research task](/docs/research-tasks/):

> Read the measurements in the linked Data folder. Compare the two conditions using the method in analysis-plan.md. Save the script, a table with units, and a note describing exclusions. Keep the input data unchanged.

Analysis scripts require the relevant local runtime and libraries. Have the assistant check those requirements and save how to rerun the work.

## Inspect the outputs

Read the script and confirm that the selected data and method match the question. Check units, sample counts, missing values, and uncertainty before accepting a result.

Keep tables and figures with their generation scripts. The [figure-prep skill](/docs/skills/) can help turn a checked result into a figure for the manuscript.

## Write the result

Ask the assistant to draft a results passage using only the reviewed outputs. Inspect each number against the table and keep the method distinct from the interpretation.

Compile the document and review the complete table or figure in the PDF, including labels and caption.