Make sense of it

Explanatory Inference

Find out why the number moved.

Explanatory Inference traces an observed outcome — a drop, a spike, a defect, an anomaly — back to its most likely causes. It ranks contributing factors by their explanatory weight and surfaces supporting evidence per cause. Reach for it when a stakeholder asks 'why did X happen?' and you need something more rigorous than manual hypothesis testing.

Shape

Operational dimensions

Requires approval

Each output waits on a human decision.

On demand

Fires when a user asks.

High data gravity

Owns a system-of-record; expensive to migrate.

Read-only inbound

Consumes external data; does not write back.

Inputs

observed outcome or anomaly (metric value, event, defect record)
candidate driver or factor space (feature set, causal model, factor hierarchy)
supporting data covering the relevant time window and segments
optional: prior attribution runs for comparison

Outputs

ranked cause attribution list with contribution magnitudes
supporting evidence per attributed cause
confidence or certainty signal per attributed factor
optional: counterfactual estimate (what the outcome would have been without factor X)

Mechanism

Identifies the most likely cause(s) or contributing factor(s) for an observed outcome / anomaly / event by attributing it to upstream drivers from a candidate explanation space.

Why this is a primitive

Cannot be decomposed: the attribute-outcome-to-cause operation is a single inference shape distinct from classification (which labels present state, not causes), prediction (which projects forward, not backward), and aggregation (which describes, doesn't attribute). The drivers-discovery / root-cause / diagnostic primitive is the backward-causal counterpart of predictive-inference.

Where it shows up

EdTech operator — completion rate drops 12% in one cohort; explanatory inference identifies that the drop is concentrated in the third module and correlates with a content update deployed that week

SaaS company — MRR growth slows; attribution surfaces that the driver is a 30% decline in expansion revenue from the mid-market segment, not new customer acquisition

Manufacturing plant — defect rate spikes on one production line; root-cause analysis ranks machine calibration drift as the primary driver over material batch variation

Retail chain — a specific store underperforms its forecast by 18%; diagnostic inference attributes most of the gap to a local competitor promotional event and one understaffed shift

Related primitives

Composes with

Classificatory Inference →Workflow Routing & Gating →Stateful Object Lifecycle →Descriptive Aggregation →

Alternative to

Classificatory Inference →