# How cubes work
URL: https://docs.valsight.ai/how-cubes-work/
Description: Every node in a Valsight model holds its values as a cube, a multi-dimensional data structure where each value is identified by its exact position across al…
Every node in a Valsight model holds its values as a cube, a multi-dimensional data structure where each value is identified by its exact position across all relevant dimensions. Understanding what a cube is explains how functions behave, why operations produce the results they do, and what “maximum cube size” means in practice.

## The core idea

[How dimensions work](/how-dimensions-work/) explains how Valsight organizes data across dimensions. A cube is the actual structure that holds those values for each node.

In a spreadsheet, data lives in rows and columns. Adding a third axis requires a new sheet, a lookup formula, or a pivot, and the structure fragments as complexity grows. In Valsight, all values for a node exist inside a single cube: a multi-dimensional space where each axis is a dimension. A single value is located by its exact position across all dimensions at once.

The flat table and the cube hold exactly the same data, just organized differently. A row like “Year = 2025, Region = EMEA, Product = Truck, Fixed Cost = 40 €” becomes a single point inside the cube, located at the intersection of those three axes.

![Valsight\_Cube.PNG](./images/7007567911.png)

This is not only a visual metaphor: it is the actual structure of every node’s output. When a formula calculates a result, the result is a cube. When a function filters, reshapes, or expands data, it operates on a cube and returns a cube with a defined shape. When the system warns about “maximum cube size,” it refers to the number of intersections in that result.

The geometric name is intentional: with two dimensions, the structure is a flat table. With three, it becomes a cube. With more, the same principles extend beyond physical visualization.

## Key components

* **Dimension**: One axis of the cube (e.g. Time, Region, Product). See [How dimensions work](/how-dimensions-work/) for levels, hierarchies, and structural rules.
* **Intersection**: A single point in the cube, defined by one level value per included dimension. A value lives at exactly one intersection.
* **Cube size**: The total number of intersections in a cube. In the worst case this equals the cross product of all level members across all dimensions; in practice, most real-world cubes are sparse.

## Why this matters

Because every node produces a cube, Valsight can apply operations consistently across any combination of dimensions, without requiring separate formulas or sheets for each view.

This enables:

* **Consistent aggregation**: Values at any level can be rolled up without rebuilding logic. The hierarchy encodes the relationship once.
* **Predictable alignment**: When two nodes are combined, alignment is always determined by their shared dimensions. There are no hidden joins or implicit matches.
* **Defined function behavior**: Every function in Valsight has a defined effect on the cube’s shape. FILTER reduces it. ROLLUP aggregates it. IS\_NA and EXPAND can fully expand it. Knowing a function’s shape behavior is central to using it correctly.
* **Diagnosable results**: Most unexpected outputs trace back to a shape mismatch: the result is too aggregated, too granular, or unexpectedly expanded. Thinking in cubes makes these issues identifiable.

Understanding cube size also explains a class of performance constraints. Some functions, in particular IS\_NA and EXPAND, create a fully expanded cube, generating a value for every possible intersection. When dimensions are large, this can produce a result that exceeds the configured maximum cube size. This is a structural consequence of the data model, not a bug.

## Design principles

* Keep cubes as sparse as possible. Real-world models rarely need full cross products; prefer functions that preserve sparsity over those that expand it.
* Understand the output shape of each function before using it. Every function either preserves, reduces, or expands the cube in a specific, predictable way.
* When diagnosing unexpected results, check the shape first: is the output missing dimensions it should have, aggregated too broadly, or larger than expected?
* A cube’s size grows multiplicatively with dimension members. Adding a new level or expanding a dimension can have significant downstream effects on calculations that reference it.

## Example

The diagram above shows a **Fixed Costs** node with three dimensions: Year, Region, and Product.

The value at the intersection Year = 2025, Region = EMEA, Product = Truck is 40 €. The same cube also holds the value at every other intersection: EMEA / Car = 175 €, Americas / Truck = 95 €, Americas / Car = 320 €, all within the 2025 slice. The 2024 slice holds a separate set of values at the same region and product intersections (45 €, 190 €, 105 €, 360 € respectively).

A formula combining this cube with another node (for example, a volume node with the same dimensions) would align automatically along the shared axes. The result is a new cube with the same shape, unless a reshaping operation changes it.

If Fixed Costs is filtered to EMEA only using FILTER, the result is a smaller cube: same three dimensions, but only the EMEA intersections. If IS\_NA is applied to check for missing values, the result is a fully expanded cube covering all Year × Region × Product combinations, whether or not data originally existed at that intersection.

## Related

* [How dimensions work](/how-dimensions-work/)
* [Dimensions](/dimensions/)
* [Limitations & Data Sizing Guidelines](/limitations-data-sizing-guidelines/)