Data Lab
Reviewed
Research data with exact analyses and visible provenance.
The Data Lab combines versioned tabular data, reproducible analysis steps, charts and a conversational agent that can explain results without inventing values.
In plain terms
A reproducible data analysis preserves the input version, method, parameters and outputs needed to inspect or repeat a result.
Pin the input
Dataset versions prevent a later upload from silently changing the basis of an existing result.
Compute, then explain
Deterministic tools produce statistics and chart values; the agent interprets those outputs instead of estimating them.
Reuse the output
Move analysis results and charts into a project, figure workflow or manuscript with their source context.
A documented workflow
From intent to inspectable output.
- 01
Upload and inspect
Add a tabular dataset, review columns, missingness and data types, then pin the working version.
- 02
Choose an analysis
Ask in plain language or select a supported statistical workflow and its parameters.
- 03
Review exact outputs
Inspect computed values, assumptions, warnings and charts before accepting an interpretation.
- 04
Connect the result
Save the result to a research project or use it as grounded material for a figure or manuscript section.
Method and boundaries
The model does not calculate by intuition
Language models are useful for choosing and explaining an analysis, but they should not be trusted to improvise numeric results. The Data Lab delegates computation to deterministic analysis tools and returns their outputs to the agent for explanation.
Every saved result records the dataset version and analysis context. This helps distinguish a reproducible output from a conversational suggestion.
- Dataset versions and column profiles
- Descriptive and inferential analysis workflows
- Meta-analysis with forest and funnel plots
- Exact chart values available for inspection
A bridge between results and writing
A result can be discussed in the Data Lab, visualised in the Visual Lab and cited in the Manuscript Studio without becoming an anonymous screenshot. The shared project context preserves where it came from.
Statistical validity still depends on study design, data quality and assumptions. The agent can surface caveats, but the researcher remains responsible for selecting and reporting the method.
Questions, answered.
- Does the AI generate the numbers?
- No. Numeric results are produced by deterministic analysis tools. The agent can explain the output and help choose an appropriate workflow.
- What happens when the dataset changes?
- Datasets are versioned so an existing analysis can remain tied to the input on which it was run.
- Can results be used in a manuscript?
- Saved results and charts can be connected to a project and brought into the manuscript workflow with their provenance.
Continue through the workspace