Insights

Analysis produces conclusions, and conclusions usually die in a chat log or a slide deck. Insights keep them alive in the workspace: a short claim — "Core HR captures 62% of tracked funding" — pinned to the query or dataset that demonstrates it. The claim and its evidence travel together.

Recording and finding insights

Every query and dataset page has an insights drawer: open it to read the claims recorded against that evidence, or add one yourself. The Insights page indexes every claim in the workspace, filterable by status, so the state of your analysis is one screen — not scattered across documents.

The lifecycle

Each insight is open, validated, or rejected:

• Validate ("holds") — you checked the claim against the current data and it stands.
• Reject ("doesn't hold") — the claim is wrong or no longer true; it stays visible, dimmed, as a record of what was considered.
• Reopen — put a decided claim back on the table when circumstances change.

Staleness

When the data or SQL underneath an insight changes — a dataset gets a new version, a transformation re-runs, a query is edited — the insight is flagged stale. Stale doesn't mean wrong; it means unverified against what's there now. Re-check the evidence and mark it "still holds" or reject it. Analysis stops silently rotting the moment the numbers move.

Faster with an agent — agents record insights as they analyze: ask for "analyze this dataset and record your conclusions as insights", and each finding lands with its evidence attached. Your job shifts from producing claims to reviewing and validating them — the judgement stays yours.