---
title: Funnel Analysis
published: true
index: 60
tier: free
---

# Collect evidence for funnel Checks

Use this procedure only for selected funnel/drop-off or commerce-event-integrity Checks. The Checks own the evaluated step, required source, applicability, evidence and outcomes. The [integration reference](../tools/integrations.md) owns target binding, read access, privacy and default windows.

Reuse compatible supplied `$report-shopify-funnel` evidence with its exact source, window, segment and denominator. A separate five-rate report or assigned-cohort readout belongs to that skill; do not require it, import its catalog or generate it automatically during an audit. Supplied evidence must still satisfy the selected Check's scope and integrity requirements.

## Collect a comparable step population

1. Verify the selected source's event/outcome definitions and integrity before interpreting the step. Use only pre-existing purchase evidence; never generate a purchase event. Evaluate a selected commerce-event-integrity Check under its own criteria when measurement is unreliable.
2. Preserve the step's ordered population, eligible denominator, counts, rates, current/comparison windows, source timezone and material segments. Unrelated raw event counts cannot establish progression. Follow a Check's explicit source requirements; do not silently substitute Shopify for required GA4, or require both for a Check that allows either.
3. Compare like scopes and challenge traffic/device/market mix, promotions, price, inventory, seasonality, consent, tracking and outages. Sparse or contradictory evidence remains inconclusive. Record data-quality blockers only against the decisions they prevent.
4. Use the material persistent gap to choose affected and comparison browser paths. Collect direct state evidence through [browser testing](browser-testing.md) and localize a reproducible cause with [debugging](debugging.md). Analytics chooses where to inspect; it does not prove an interface cause.
5. Return evidence to the selected Checks. Reuse shared observations without merging their decisions; validated reach may adjust an existing Issue's priority without creating a duplicate funnel action.

## Recommend a test

Use the [reporting contract's Issue gate](../core/reporting-contract.md#issue-inclusion-and-quality-gate) for experiment content, category ownership and supported-action decisions. Keep an unproven diagnosis in Check evidence rather than inventing a winner, uplift, sample size or new action. Preserve supplied owner decisions and compatible prior learning under the operating contract.
