DocsFeature guides

Actions

The recommendation engine's output as a persistent kanban board whose cards score themselves by priority and auto-close when the data says the gap is fixed.

Actions turns the deterministic recommendation engine into a persistent kanban board. Cards move through Suggested, Planned, In progress, and Done (with a separate dismissed drawer), and they auto-close — marked "verified by data" — when the engine stops emitting them.

What it does

The recommendation engine reads your real metrics and emits a scored list of the highest-leverage things to fix. Actions persists that list as a board you can work, so a recommendation you have started isn't lost the next time the engine runs — and a card resolves itself when the underlying gap closes.

How to use it

Note

Batch actions are on the way. Today the board supports per-card moves plus a bulk restore of dismissed cards; there is no multi-select bulk edit yet. This page will be updated when batch actions ship.

Open Actions and work the board:

  • Move a card by drag-and-drop or the per-card dropdown. Moves are optimistic; in demo mode they are local-only and not saved.
  • Filter with the severity chips, which double as filters.
  • Scope the view with deep links: ?prompt= and ?source= open a filtered board.
  • Each card offers Brief, Evidence, Source, and Track.

Content briefs live at /actions/[promptId]/brief. A brief is generated by a language model grounded in your real answers, citations, and knowledge base, and is cached once per prompt. Regenerating deletes that cache and draws from your monthly draft allowance — see Plans and quotas.

How it's computed

Fourteen pure rules score each recommendation. Priority = round(100 × impact × confidence). Severity follows from priority: high when at least 66, medium when at least 33, otherwise low. Confidence saturates at five runs, so a recommendation can't out-rank the sample behind it.

The "verified by data" sync reconciles the board against each fresh engine run. It:

  • inserts new recommendations,
  • refreshes snapshots that changed,
  • marks open cards resolved when the engine no longer emits them, and
  • un-resolves a card if the gap returns, unless you have already closed it manually.

An open card is one whose status is not done or dismissed and which is not resolved — the same definition the Portfolio uses for its open-actions count.

Limits

Actions is available on every plan. It is presented as a Growth feature in marketing, but it is not gated at runtime.

  • Content — where briefs and drafts are generated and metered.
  • Fan-outs — source of the fan-out-opportunity recommendation.
  • Portfolio — rolls up open actions per brand.