Эта страница на английском

Мы её пока не перевели. Остальная часть сайта доступна на вашем языке.

Product documentation

Bind or unbind an analysis model

When the Analysis Model section is available, a dataset can use a compatible model instead of its own editable knowledge for analysis.

Before you bind

You need corresponding access. DataFact offers only models discovered as compatible with the dataset's database connection and schema scope. A model may therefore exist without appearing in the selector.

Bind a model

  1. Open the dataset and select Analysis Model.
  2. Select a model from the available candidates.
  3. Select Bind.
  4. Wait for the success message and verify that the model card is marked as in use.

The card can show the model name, type, description, and current build version when those values are available. View Model opens the model details.

Replace a binding

  1. On the bound model card, select Rebind.
  2. Select another compatible model.
  3. Select Bind.

Binding and replacement take effect without a separate confirmation dialog. Verify the model name after success.

Unbind a model

  1. Select Unbind on the bound model card.
  2. Read the confirmation dialog.
  3. Confirm the action.

Unbinding removes the model association. For an eligible dataset version, dataset-level knowledge becomes editable again. DataFact does not promise to change the dataset's version when unbinding.

Exceptional states

  • No candidates: no compatible model is currently discoverable. Create or update a compatible model, then retry.
  • Bound model cannot be resolved: the association exists, but model details could not be loaded. Rebind or retry later.
  • Bound model was deleted: the card identifies the deleted state. Rebind to another model or unbind.
  • Request error: use the retry action when offered; do not assume the binding changed unless success is confirmed.

Important effect

For dataset versions that keep dataset-level knowledge visible, binding makes that knowledge read-only. Unbound analysis can fall back to the dataset's own knowledge behavior; the exact answer still depends on the question, source data, and runtime execution.

Last verified: 2026-08-18Initial draft complete