Models
One agreed set of joins and calculations, so every dashboard in a space works its numbers out the same way.
A model is your space's description of its own data: which tables relate to which, how, and what your named figures actually mean. Dashboards read it instead of each working the same thing out for themselves.
Without one, every dashboard is on its own. It infers how two tables line up, and it writes its own arithmetic for "active customers" or "net revenue". Two dashboards built a week apart can both be reasonable and still disagree. A model is where you settle it once.
Find models at the foot of the sidebar in Data.
What is in one
| Tables | Datasets you have put in the model, taken as they stand. Adding one does not copy or move anything. |
| Joins | How two tables relate: the columns, the cardinality, and which way a filter travels. |
| Calculations | Named formulas, defined once and available to every dashboard reading the model. |
Create one
Open the tray
In Data, open Models at the foot of the sidebar. Drag the bar to resize it, click to expand or collapse.
Add a model
Use the add button on the bar and name it. A space can hold as many as it needs, one per subject area or one for everything.
Put tables in it
Drag datasets onto the canvas from the list on the left. Multi-select and drag them together, or use Add to model in the selection bar.
Join them
Flitch looks for joins as tables arrive and holds anything it cannot prove in a review queue. Draw the rest yourself by dragging between two tables. See Tables and joins.
Point a dashboard at it
A model changes nothing until a dashboard reads it. See Models and dashboards.
Sections
Tables and joins
The canvas, how joins are found, and which way filters travel.
Calculations
Named formulas, the grammar, and importing between models.
Models and dashboards
Binding a dashboard, coverage, publishing, and generation.
The default model
One model in a space is the default, marked Default beside its name. The Auto switch on the tray bar starts every new dashboard on it, with its tables already picked. Switched off, each dashboard chooses for itself.
Change which one it is from a model's menu with Set as default. The same menu duplicates, renames, and deletes.
When you do not need one
- A single dataset, or a single dashboard. There is nothing for two dashboards to disagree about yet.
- Tables that never get read together. A model describes relationships, and unrelated data has none.
A dashboard with no model works its own numbers out, the way it always has. That is a supported state, not an unfinished one.
Models never write to your sources. Joins and calculations are read at the moment a dashboard fetches its rows, so editing one reaches every dashboard on its next fetch, with nothing rebuilt and nothing copied into dashboard code.
Credits
The canvas costs nothing. Joins are found by comparing column names and sampled values rather than by an AI, so detection, editing and running a calculation are all free.
Asking the Data assistant to build part of a model for you costs credits like any other question. See Billing & Usage.