← ClaudeAtlas

ga4-schema-tracking-auditorlisted

Audits GA4 event tracking for drift against a tracking plan—catches missing required parameters, data type mismatches (e.g. "49.99" instead of 49.99), and casing drift (e.g. pageLocation vs page_location) across both standard web analytics events (page_view, generate_lead, sign_up, search, login, file_download) and ecommerce events (view_item, add_to_cart, begin_checkout, purchase), including item-scoped parameters inside the items[] array. Use this whenever the user wants to QA, validate, or audit their GA4 tracking implementation, check a GA4 BigQuery export, GTM Preview dump, Data API pull, or tracking plan spreadsheet for tagging errors, investigate why GA4 reports look fragmented or incomplete, or reconcile an analytics tracking plan against production data.
samikshapattanaik-cmyk/ga4-analytics-data-quality-suite · ★ 0 · Data & Documents · score 62
Install: claude install-skill samikshapattanaik-cmyk/ga4-analytics-data-quality-suite
# GA4 Schema & Tracking Auditor This skill compares observed GA4 events against a reference tracking plan, reporting gaps according to three severity tiers: - 🔴 **CRITICAL:** A load-bearing event/parameter is missing (breaks core GA4 processing or revenue reporting), or a load-bearing parameter's item-array is missing entirely. - 🟡 **WARNING:** A parameter is present but wrong — a data type mismatch (e.g. `"49.99"` instead of `49.99`). WARNING is reserved for something actually being wrong, never for a parameter simply being absent. - 🔵 **NOTICE:** Either a recommended-but-not-load-bearing parameter is absent, naming/casing drift is fragmenting reports (e.g. `pageLocation` vs `page_location`), or an observed parameter is a known common misnomer for a real one the plan expects (e.g. `item_code` when the plan expects `item_id`). Absence of anything non-critical is deliberately kept at this lower tier so the report stays focused on real problems instead of nagging about every optional field a legitimate implementation chose to skip. All parsing, unnesting, and diffing logic is executed deterministically via `scripts/validate_schema.py`, which emits two views of the same findings: - **`findings`** — granular, one entry per (event, scope, parameter, issue), for anyone who wants full detail. - **`event_reports`** — aggregated one row per distinct (event, issue-pattern) combination, each bundling every issue for that pattern together with a single ready-to-paste `dataLayer.pus