---
title: Performance Testing
published: true
index: 70
tier: free
---

# Collect performance Check evidence

Use this procedure for shared measurement setup across selected performance Checks (`technical.performance.*`). Each Check owns its required diagnostic, applicability, test steps, evidence and outcome. Do not replace that catalog with another checklist or skip selected Checks because one metric dominates.

## Reuse and measure

Reuse compatible supplied `$report-shopify-performance` results with their route, viewport, method and revision. A standalone eight-metric baseline or release comparison belongs to that skill; it is not required for this audit. Collect any additional evidence required by selected Checks using existing host tools.

For new measurements, record route, environment (live, preview or local), theme/revision, consent, geography, cache state, device/network profile, timestamp and tool version. Use representative routes and states required by the Checks. For comparative lab claims, use at least three comparable runs, retain sample values, median and variance, and never select only the best result. Keep field percentiles, device, geography and date windows separate from lab traces.

Preserve native TTFB, FCP, LCP, CLS and interaction definitions. Never subtract TTFB and relabel the result LCP; Lighthouse TBT is not INP, and a local interaction is not a real-user percentile. Missing fields, transfer counts or real-user samples are unavailable, not zero. Do not apply Theme Store submission scores to merchant storefronts.

## Attribute the selected Check's cost

Start from the affected rendered element, interaction or request and follow its timing, initiator and active owner. Use [debugging](debugging.md) for source attribution. Existing host traces, Network initiators, Lighthouse Treemap or Theme Inspector may help; availability does not make a tool mandatory when other evidence satisfies the Check.

Use the selected Check's specific procedure for images, video, fonts, Liquid, scripts, apps, delivery or checkout extensions. Confirm lint and automated leads on the active path before treating them as a customer problem. Preserve independent causes when owners, changes or retests differ. Optional strategies are not defects merely because absent.

Use the [operating contract](../core/operating-contract.md) for read-only boundaries and checkout limits, and the [reporting contract](../core/reporting-contract.md) for Issue eligibility and priority. Measurement alone does not prove a funnel cause, and publisher speed/conversion coefficients cannot establish this store's recoverable revenue or expected uplift.
