Skip to main content
While a single measure is only a fact, the Metric Tree captures how measures relate to one another, so that a top-line metric such as ARR, labor cost %, or contribution margin decomposes into the driver metrics that explain it. It is the part of the World Model that turns “what happened” into “why,” and because it is built entirely from your semantic model definitions, there is no separate file to maintain.

How measures compose

Measures connect through two kinds of edges, which differ in what they claim.

Component edges

When one measure’s expression references another with {{view.measure}}, that reference is an edge: arr depends on net_mrr, which in turn depends on total_mrr and churned_mrr, forming a tree the engine reads straight from the formulas. The operator in front of each reference (+, -, *, /) tells it whether the relationship is additive or multiplicative.

Driver edges

Where component edges are arithmetic, driver edges (declared with drivers:) encode the causal map, stating which metrics push a target up or down, in which direction, and how confident you are that the relationship is real.
Driver edges are deliberately qualitative, stating what drives what without claiming how much. A magnitude is a causal claim, and observational data cannot reliably support one, because a coefficient fitted from history absorbs everything that happened to move alongside it. Magnitudes come from the control layer instead: when a control acts on a driver, evaluation estimates the driver’s effect from the interventions themselves, and that estimate becomes the authoritative one. An edge that no control acts on either stays qualitative or carries an observational estimate that is labeled as weak.

Promotion up the hierarchy

Because the entity graph defines a parent: hierarchy (sale → store → company → market), a measure defined once at the sale grain automatically promotes upward, so markets.net_sales is derived by rolling up sales.net_sales rather than being redefined. The tree therefore captures not only how metrics compose but also every level of the business at which they apply.

Explore it in the IDE

The Metric Tree is a tab in the Semantic Model IDE, where it renders your measures as an interactive graph laid out from the top down:
  • Composite nodes (the metrics built from others) are highlighted, driver nodes are tinted amber, and leaf inputs are muted.
  • Component edges are drawn solid, while driver edges are dashed and animated, and appear fainter when confidence is low.
  • Clicking a measure selects it, dimming unrelated edges and opening a side panel with the measure’s definition and its ranked drivers, sorted by influence.

What you can do with it

Because the tree knows how every metric is built, it powers four operations over your metrics:

Sensitivity

Rank the drivers of any measure by how much they move it, using the best available magnitude on each edge: control-learned where a control acts, labeled observational estimates elsewhere.

Predict

What-if propagation: change a driver by some delta and see the estimated impact ripple up to the top-line metric, carrying each edge’s confidence.

Explain (RCA)

Decompose a metric’s period-over-period change across its drivers and the segments that concentrate it, with warnings for Simpson’s paradox and offsetting effects. Anomalies surface in the Insights Inbox with an AI-authored root cause. Available today.

Opportunity

Size the gap between each segment and its benchmark, and propagate the top opportunity through the tree. Available today. See Opportunity sizing.
For example, an explain on labor_cost_pct for a location might read:

Next steps

Measures

Define the measures, and their drivers, that the tree composes

Entities & relationships

The hierarchy measures promote across

World Model

How the metric tree fits with the semantic model and entity graph