aggregate

aggregate turns row-level data into named summaries. It emits one row per group, or one global summary row when groupBy: [].

This is the node to reach for when the output columns are the story: counts by site, mean score by arm, maximum date per account, or a compact table for a report.

Aggregation contract

Field	Required	Notes
inputs	yes	Exactly one upstream table.
groupBy	yes	Array of expressions. Empty array means one global summary row.
metrics	yes	One or more alias expressions like "[mean_age] = [age].mean()".

Designing metrics

• Each metric should be an alias expression, for example "[mean_score] = [score].mean()".
• Keep metric names report-ready. Anonymous or machine-looking aliases make the resulting table harder to review.
• Common reducers include .sum(), .mean(), .count(), .min(), .max(), .n_unique(), and .distinct().

What changes

default contains the group keys plus metric columns. The row count usually collapses, so output shape is the first thing to inspect.

For more complex windowed reductions or custom statistics, move to a Python, R, JavaScript, or SQL node.

Example

- id: by_site
  kind: aggregate
  inputs: [data]                 # length 1
  groupBy: ["[site]"]
  metrics:
    - "[mean_score] = [score].mean()"
    - "[n] = [score].count()"

Related

• Expression language - group and metric expressions