In a dataset, the grain is the level at which a fact exists: one transaction, one customer, one day, or one region.

Before analysis begins, decide what one row means.

This grain has a color, a size, and a shape. Identity remains attached.

At fine resolution, all three are still visible.

The population holds 72 records. Our grain is one of them — still visible, still traceable.

The coral outline follows the same record through the entire story.

Grains move toward others of the same color. Color is preserved because color defines the groups.

Identity, exact size, and shape are no longer the focus.

Seventy-two input rows become four output rows: one row per color group. Each summary grain's size encodes the combined sizes of its group.

The original combinations of color, shape, size, and identity cannot be reconstructed from these four rows.

The color groups dissolve. Before grouping by shape, every grain returns to the field.

The data is free — for a moment.

Group by shape instead, and shape survives. Color becomes hidden inside each summary, while sizes are combined.

A different aggregation preserves a different truth.

Before a new aggregation can begin, the old one breaks apart. Each grain returns to the field.

The data is free — for a moment.

Low, medium, and high size bands remain visible. Exact sizes, colors, shapes, and identities collapse into three summaries.

All records become one total. Fast to compare, easy to communicate — but it cannot explain who, what, or why.

The grain is now: one row for the entire population.

Fine and coarse grains answer different questions. The mistake is not aggregation — it is accidental aggregation.

Define the question. Preserve the attributes that matter. Combine only what the decision does not need.

The right grain contains enough detail — and no accidental loss.