Live and frozen BI artifacts

BI artifacts have two different jobs.

Some artifacts are monitors. They should stay live because the user wants the latest answer every time they open them. A weekly risk view, freshness check, or management dashboard belongs in this category.

Other artifacts are evidence. They should freeze because they supported a decision at a specific moment. A board pack, investor update, credit-policy review, or incident note needs to preserve what the team saw at the time.

Treating both as "charts" hides the product distinction. The important choice is whether the artifact is meant to monitor a changing state or preserve a decision state.

Product implications

Live artifacts need a saved query spec, verification path, refresh state, and failure behavior. A live chart that silently refreshes from a changed definition is dangerous.

Frozen artifacts need provenance. The user should be able to recover the snapshot date, filters, metric definitions, row count, and caveats that supported the artifact.

When BI becomes interrogable, this distinction matters more. A chat answer can become a chart, a report, or a recurring check. The system has to know whether that output should stay current or stay faithful.

Notto example

Notto AI Agent supports frozen artifacts by default. Live chart and table artifacts require saved SQL plus the structured query intent, and refresh goes back through the verified SQL path.

That is a small implementation detail with a large product meaning: current views and decision evidence should not collapse into the same thing.