Analyze and troubleshoot with Statbot

Use Statbot to inspect Statsig configurations, interpret experiment results, and investigate missing data or unexpected targeting.

Early Access

This feature is in Early Access. During this time, aspects of the functionality may still be developed, and this documentation may not always be up to date. If you have any questions, contact Statsig Support.

Ask Statbot about a Statsig resource to inspect its configuration, understand available results, or investigate a problem. Statbot combines page context with project data and official documentation. Open the resource in the Console to check its answer against the underlying setup or results.

Prerequisites

Enable Statbot for your project.
Obtain MCP read access and permission to read the resources you want to inspect.

Inspect a configuration

Open the gate, experiment, dynamic config, or metric you want to inspect.
Select Statbot and ask a specific question, such as "Explain this gate's targeting rules."
Ask follow-up questions about a rule, group, metric, or parameter.
Open the resource links in the answer to verify the details.

To find a resource first, ask "List my active experiments" or "Find the gates I created." Include an exact name or ID when several resources have similar names.

For a dynamic config, include the user attributes and environment relevant to your question. For example: "Given country Canada, explain which rule in homepage_preferences matches and what values it returns."

Interpret experiment results

Open the experiment's results in the Console.
Ask "Compare Treatment with Control for the primary metrics. Include lift, statistical evidence, and guardrails."
Specify the date range, groups, or metric if the answer needs a narrower comparison.
Check Statbot's answer against the experiment results and diagnostics before deciding whether to ship.

An experiment that hasn't started has no experiment outcome to interpret. A running experiment can also lack completed results. Neither state establishes that the treatment has no impact.

A result that isn't statistically significant doesn't prove that the treatment has no effect. Check the experiment's results interpretation guidance and investigate data-quality warnings before relying on a recommendation.

Investigate missing data

Tell Statbot what you expected, what the Console shows, and the affected time range. For example:

"This experiment started yesterday, but its results aren't available. What can you verify about its status and exposures?"
"Why is this metric unavailable for this experiment? Check its definition and the experiment setup."
"What changed in this gate before its rollout increased?"

Ask Statbot to distinguish facts it retrieved from possible explanations. Follow its resource links to inspect the evidence, then use the relevant Console diagnostics to investigate anything Statbot couldn't verify.

When Statbot can't answer

Symptom	Next step
Statbot can't find a resource.	Provide its exact ID, check the selected project, and verify your read permissions.
Statbot can explain Statsig but can't read project data.	Ask an administrator to check project and role MCP access.
Results or chart values are missing from the answer.	Open the results in the Console. Access to a chart definition doesn't guarantee access to its plotted values.
The answer concerns a different resource.	Name the intended resource explicitly or start a new chat.
The answer appears incomplete or incorrect.	Ask for the source and missing details, then rate the answer with the feedback controls.

To change a configuration after investigating it, follow Create and edit with Statbot.

Was this helpful?