Attribution Window Sanity Check

Attribution Window Sanity Check

Use the shared quality bar in ../references/output-standard.md and ../references/skill-design-principles.md when those files are available.

Use this skill when

• the user shares GTM, GA4, Google Ads, Meta Pixel/CAPI, consent, click ID, data layer or CRM import evidence tied to attribution window sanity check.
• the next decision could change conversion actions, tags, triggers, event parameters, imports, consent checks or bidding signals.
• platform counts disagree with backend, CRM or debug traces, and the team needs to know which signal to trust.

Do not use this skill for broad analytics theory. Use it when tracking evidence, platform counts, debug traces or import data must be checked before an operator trusts the signal.

Required input

• GTM container export, tag list, GA4 event list, Google Ads conversion actions, Meta Pixel/CAPI setup or screenshots.
• debug traces, preview screenshots, data layer examples, click IDs, consent state and recent tracking changes.
• platform reports for the same time window so browser, server, GA4 and ad platform counts can be compared.
• what decision the user is trying to make next: verify, debug, deduplicate, import offline conversions or assess readiness.
• If an input is missing, continue with a clearly marked assumption instead of inventing data.

Analysis workflow

1. List attribution windows across Google Ads, Meta, GA4 and CRM reports.
2. Compare windows with sales cycle, consideration time, conversion lag and reporting cadence.
3. Identify mismatches that make channel comparisons unfair or misleading.
4. Check whether optimization windows match the event being optimized.
5. Recommend reporting alignment without rewriting historical performance claims.

Decision rules

• If the data does not connect to revenue, pipeline, qualified leads or conversion quality, label the recommendation as a hypothesis.
• If platform metrics and downstream data disagree, trust the downstream source for business quality and platform data for delivery mechanics.
• If the issue could be tracking, offer, audience, page or follow-up, do not collapse it into one cause without evidence.
• Do not tell the user to edit live tags, containers, consent settings or conversion actions without an approval step.

Output format

End with:

• Decision: fix / test / monitor / ask for data / do not act yet
• Approval needed: yes/no and what would change if approved
• Missing data: only the inputs that would materially change the recommendation

Practical example

User: "Here are GTM screenshots, platform counts and debug traces for attribution window sanity check. Is this signal safe to trust?"

Assistant should: use the supplied evidence, run the workflow above, produce the skill-specific rubric or diagnostic table, and stop at approval-ready recommendations.

Guardrails

• Do not make changes to live campaigns, pages, tags, containers, CRM fields or customer messages.
• Do not claim performance impact without evidence.
• Mark missing data clearly.
• Keep recommendations practical for a performance operator, founder or owner with a real advertising problem.