Star schema

Review the AI's star schema design

When the analysis finishes, the setup dialog moves to Review AI Recommendations. Nothing is saved to the plan yet. This is where you decide which dimensions and facts to keep before Kenseme creates them.

In the app: Data Model › Schemas › your source schema › Star Schema tab › New Plan › Analyze Schema

[screenshot]
The Review AI Recommendations step: the bus matrix with checkboxes, the detail panel for DimCustomer, and the Build Order below.

Read the summary

The top line reads AI recommended 8 dimensions and 3 fact tables and keeps a live count of your selection, for example 6 dims, 2 facts selected.

If some subject areas failed during analysis, a yellow Design completed with errors alert names them. The rest of the design is ready to review. Click Retry failed areas to run only those areas again.

Read the bus matrix

The Bus Matrix grid puts dimensions in rows and facts in columns.

  • Each row and column header has a checkbox, the item’s name, and its type underneath (for example SCD2, Transaction, PeriodicSnapshot).
  • A cross in a cell means that fact uses that dimension. It is bold when both are selected and grey when either one isn’t.
  • A dash means the fact doesn’t use the dimension.
  • Role-playing dimensions appear with their role, for example DimDate (OrderDate).

Everything starts selected.

Look at one item

Click an item’s name to open its detail panel below the grid. It shows:

  • The name, its type badge, and its category badge.
  • The AI’s reasoning for proposing it.
  • Grain: for a fact, for example “One row per order line”.
  • Dimension FKs: the dimensions a fact looks up.
  • Referenced by: how many selected facts use a dimension.

Keep or drop items

  • Untick a dimension or fact to leave it out of the plan.
  • When you untick a dimension that a selected fact uses, that fact gets a warning icon (Prompts will need regeneration) and the summary shows n need prompt refresh. The fact’s design still mentions the dropped dimension. The flag carries over to the plan so you can refresh the fact’s prompts before you build it. Tick the dimension again and the flag clears.
  • When you untick every fact that uses a junk dimension, the junk dimension is unticked for you.

Create Plan stays disabled until at least one fact is selected.

Refine items with extra guidance

If some items are close but not right, ask the AI to redesign only those.

  1. Make sure only the items you want redesigned are ticked.
  2. Click Refine selected with guidance….
  3. In the Refine n checked item(s) panel, type what should change, for example “Track customer segment history as SCD2 and add a loyalty tier attribute.”
  4. Click Start refine.

The AI redesigns only the ticked items. Everything else stays as it is. Refining counts toward your AI usage.

Check the build order

The Build Order section shows the order the selected items will be built in, for example DimDate → DimCustomer → DimProduct → FactSales. Dimensions come before the facts that use them. The order updates as you tick and untick items.

Create the plan

  1. Click Create Plan. The dialog shows Creating plan items....
  2. Kenseme saves the selected items and their dimension lookups, and creates a Pipeline named after the plan with one table mapping per dimension and fact.
  3. A toast confirms Plan created with the number of items added.
  4. If anything needs your attention, a Plan created with warnings dialog lists it. Examples: an SCD type 3 or 6 dimension that was downgraded to a type the Pipeline supports, or a fact foreign key that didn’t match any dimension.
  5. The plan dashboard opens.

Click Cancel instead to close the dialog without creating any items.

Note: The plan stays editable. You can add items by hand, edit dimension lookups in the plan’s bus matrix, and change settings later.