--- title: Landing Trends and Experiment Cohorts published: true index: 10 --- # Landing-page trends and experiment cohorts ## Comparable landing-page evidence Use the [funnel workflow](funnel.md) for source binding, event integrity, rate definitions and output. Read available aggregate Shopify or GA4 data independently. Record the actual report/query, source identity, extraction time, store/property timezone, complete current and comparison windows, filters, session denominators, outcome counts and reporting delays. Do not imply that a limited query covers every landing page. Normalize landing routes using actual hostnames and redirects. Strip tracking parameters and redact private query values while preserving functional market, locale and variant distinctions. State how unknown routes, consent, bots and internal traffic affect coverage. Compare each route with its compatible prior period and, when useful, matched routes of the same intent. Account for device/channel/market mix, promotions, inventory, price, seasonality and theme or instrumentation changes. Do not sum overlapping session populations or compare a product landing with an editorial goal as if their intent were equal. Keep every numerator, denominator and comparator within one source and definition. A GA4 user conversion rate, an event count and a Shopify session conversion rate are different measures. Use the declared session rates only when that source supplies the required session outcomes. Preserve numerator and denominator alongside each rate; a zero denominator or missing outcome is unavailable, not zero. Show absolute percentage-point change and counts; discuss sparse or unstable samples without inventing a universal benchmark or significance threshold. Report the material improving, declining and inconclusive cohorts, explaining their reach and uncertainty. A period comparison is an observation, not a randomized test. Reuse compatible supplied audit evidence and existing Issue IDs without rerunning or duplicating their actions. This report owns landing-page measurement and trends; it does not require a prior audit. Direct storefront observations are required before proposing a cause, fix or falsifiable experiment. Otherwise provide a next investigation in Check evidence or a roadmap handoff. For optional organic acquisition context, use the [Search Console procedure](../tools/search-console.md). Compare normalized landing pages with separately defined Shopify or GA4 organic segments, preserving each source's timezone, dates, attribution, and denominator. Search clicks and analytics sessions are different populations; URL matching alone cannot join individual users or establish experiment assignment. Keep acquisition metrics and uncertain explanations in the referencing Check evidence without adding Data keys, dimensions, or catalog Checks. ## Supplied experiment Resolve the store, native experiment ID/link, baseline and treatment theme IDs/revisions, affected audience and routes, actual assignment method, exposure window and allocation. Read the agreed primary metric, guardrails, sufficient-evidence and stopping criteria from the brief or owner policy. Without trustworthy assignment, return an inconclusive experiment readout; URL parameters, preview cookies, theme IDs, acquisition campaigns and before/after dates alone do not establish randomized cohorts. Use native experiment results or an authorized source with explicit, validated assignment and compatible denominators. State whether measurement is by visitor or session and whether it is intention-to-treat or exposed-only. Native visitor metrics must not be relabeled as session rates: retain them with their definitions in Check evidence, and emit the five declared Data keys only for supported session metrics. Keep control and treatment counts, effect estimate, uncertainty and native platform status attributable to the actual source. Do not pool native visitor results with GA4 sessions or infer missing cohorts from store totals. Check delayed purchases, consent/attribution loss, uneven or changing allocation, contamination, theme edits during the test, conflicting experiments, instrumentation changes and segment imbalance before interpretation. Keep incomplete windows, sparse cohorts, unavailable guardrails or unmet decision criteria inconclusive. Report a breached guardrail even when the primary metric improves. A positive snapshot or a platform suggestion alone does not establish a winning change. This report reads evidence and proposes the next action; it never launches, stops or promotes a test. Hand the actual readout and limitations to the owner's roadmap or winner-evaluation workflow. Where the host is missing a read capability, name the missing operation; do not create tracking, change allocation or substitute synthetic preview results. ## Existing report interface Keep the five rate keys and exactly the existing `window`, `segment` and `source` dimension keys. Use a stable, explicit `segment` value for the cohort, such as `landing=/collections/shoes;device=mobile` or `experiment=;variant=control;landing=/collections/shoes`. Preserve the aggregate `all` rows. A cohort label records selection; it never proves assignment. Keep definitions, sample counts, comparisons, guardrails and inconclusive readouts in the referencing Checks/evidence when they do not fit a declared Data key. Do not invent a new result schema, Data key, dimension or extra evaluative catalog Check. Complete each of the five selected Checks individually. Keep available Data even if its diagnosis is blocked. A missing source, custom progression event or experiment assignment blocks only the dependent decision. Existing non-experiment funnel reports continue without an experiment requirement. Use the existing Markdown Data, Issues, Checks and Limitations sections or the host-supplied Glue schema; no Summary or Findings. Feed one deduplicated, evidence-linked next action to the roadmap when warranted, keeping neutral trends and unresolved causes in Checks. For relevant speed or visual changes, reuse performance reporting (`$report-shopify-performance`) and visual drift reporting (`$report-shopify-drift`). Synthetic previews cannot supply funnel or experiment outcomes.