# Comparison guide

## Evidence-strength labels

- `controlled`: assignment or design credibly isolates the intervention.
- `adjusted`: important differences are measured and explicitly adjusted.
- `directional`: the periods are broadly comparable, but alternative explanations remain.
- `not-comparable`: population, workflow, or measurement changed enough to make the numerical contrast misleading.

## Required guardrails

At least one quality or acceptance measure must accompany speed or volume. Depending on the workflow this may be reopen rate, defect rate, escalation, customer correction, approval rate, exception rate, policy breach, or downstream rejection.

Do not translate elapsed time reduction into labor savings unless released capacity, staffing change, or redeployed work is observed. Do not translate more completed items into business value without a named downstream result.