pureinn.docs
Phase Cross-phaseAgent mode: neverv1.0.0

pm-root-cause

Diagnostic engine for in-flight anomalies - drills to the real root cause, not the first plausible one

/pm-root-cause

When to use

When something live behaves unexpectedly: a metric dropped, churn spiked, a feature isn't adopted, conversion fell, tickets are rising. Cross-phase - run whenever an anomaly needs structured diagnosis, at any point in the product lifecycle.

Agent mode is not supported. The value is the live investigative dialogue - hypothesis drill-down requires interactive back-and-forth. Invoking with --agent will warn once and proceed interactively.


What it produces

A structured root cause investigation with:

  1. Reality check - is this a real anomaly or a measurement artifact (data quality, segment definition, seasonality, instrumentation bug)?
  2. Concentration analysis - where does the anomaly concentrate? Segment / funnel stage / cohort / geography / feature / device?
  3. Change inventory - what changed (product, marketing, competition, external event) in the relevant time window?
  4. Candidate cause differential - causes ranked by evidence strength, across categories: product, data/measurement, market, operational, team
  5. 5-Whys drill - structured iterative drill on the highest-confidence candidate
  6. Evidence vs. guess separation - what is confirmed vs. inferred vs. assumed
  7. Cheapest next test - the lowest-cost experiment to confirm or reject the root cause hypothesis
  8. Testable hypotheses - formatted for input into pm-hypotheses

Built-in diagnostic methods

  • 5-Whys drill
  • Fishbone (Ishikawa) categorization
  • Anomaly-to-cause differential library
  • Measurement trap catalogue (common instrumentation errors)
  • Cognitive bias catalogue (availability, confirmation, recency bias checks)

How to invoke

/pm-root-cause    # interactive only - --agent not supported

Describe the symptom: what metric, what change, over what time period, compared to what baseline. The skill takes it from there - you do not need a hypothesis before starting.


Dependencies

Recommended (provides context):

  • pm-kpis - KPI framework helps scope the anomaly and confirm the metric definition
  • pm-personas - segment breakdown informs where the anomaly concentrates
  • pm-hypotheses - testable hypotheses from this session feed directly into a new experiment plan

Produces for:

  • pm-hypotheses - root cause hypotheses become entries in the Hypothesis Register for validation
  • pm-prioritize - confirmed root cause may change feature priority

Related skills: pm-hypotheses, pm-problem-validation, pm-kpis, pm-personas, pm-prioritize