Client asset pack
acceptance report

1

Symptom. A clean cache HIT on the billiard-table landing page serves one visitor's UTM parameters and gclid inside the three rendered CF7 form action URLs to the next visitor who hits the same cache entry.

2

Root cause. The production cache layer, owner not yet identified, caches query-derived personalization fields as part of the static HTML instead of stripping them before the page is cached.

3

Fix. Exclude query-derived click IDs and UTM values from cached HTML. Generate attribution fields per request after the cache layer. Never cache a personalized form action.

4

Who. Site or cache owner, currently unassigned.

1

Symptom. The Google Ads "Submit lead form" conversion action is a page-load trigger on a URL that returns 404. A separate "Purchase" action sits on a URL that returns 200 and is armed but silent.

2

Root cause. Form start stayed Primary and counted Every in Google Ads while the real submission-confirmation URL was never wired to a working conversion action, so both spend and bidding optimize on a page load, not a submission.

3

Fix. Generate one lead_id on server-accepted CF7 success, map it to one canonical GTM and GA4 event, then make the deduplicated submitted-lead action Primary, count once, and move form start and the dead Purchase action to Secondary or retire them.

4

Who. Site or GTM owner for the event, Google Ads account owner for the conversion action change. Both currently unassigned.

1

Symptom. crm.source_column is null on every enquiry row in the form register. No submitted enquiry can be traced back to Google or Meta spend.

2

Root cause. The form register headers never included source, campaign, click-ID or landing-page fields, so a submission lands with no attribution even where a click ID was technically available upstream.

3

Fix. Add lead_id, source, medium, campaign, content, term, landing URL and every click ID to the Sheet, populated from the fields the EDIT-02 payload already captures at submission time.

4

Who. Sales or CRM owner for the sheet schema, coordinated with the site owner who ships EDIT-02. Both currently unassigned.

1

Symptom. The Google Ads conversion_value field carries the raw conversion count tagged with unit "huf", producing a ROAS of 0.0005 that reads as a real return-on-spend figure.

2

Root cause. No verified revenue figure exists anywhere in the stack. The CRM schema has quote and payment fields, but no filled Fusion outcome value, so the count was left in the value field as a placeholder.

3

Fix. Remove the forint-labelled value from the conversion action now. Report spend, clicks and enquiries only, never a computed ROAS, until offline enhanced conversions import a real quote or purchase value from the CRM.

4

Who. Ads account owner for the immediate removal. Sales or CRM owner for the eventual real-value import. Both currently unassigned.

1

Symptom. One report run pulls Google Ads and Meta Ads on 2026-07-25 to 08-23, BigQuery on 2026-07-24 to 08-22, the CRM sheet on 2026-07-26 to 08-24, and work evidence on 2026-06-23 to 08-23.

2

Root cause. Each collector pulls its own trailing window independently, driven by per-source lag and API history limits, and the report assembler merges them without normalizing to one canonical window.

3

Fix. Pick one canonical window per report run. Have every collector clip or pad to it, or return UNAVAILABLE for that source. Treat source_health.range as a gate a report cannot pass with mismatched ranges.

4

Who. The reporting-pipeline owner, the same code line that writes report_data.json.

1

What it is. Four registries plus one A/Z test ledger, rebuilt every month from the raw Meta, Google and leads-sheet pull by a single script. It turns the month-end report into a decision list instead of a description.

2

What it answers. Which creatives to change, alter or keep. Which copy to cut, alter or keep. Which audiences to exclude or double down on. What was tested and what the test decided.

3

What it never does. No Meta or Google writes. It reads, ranks and recommends. Changes are applied by hand in Ads Manager after the recommendation is accepted.

4

How to re-run. Pull the new month with the three scripts in qa/data/, then run _registry.py. The four CSVs are rewritten in place next to docs/fusion-pulse/README.md.

Creatives

• Keep at or below the median CPC, or within 20 percent above it.
• Alter 20 to 50 percent above the median, new colourway or first frame, same concept.
• Change more than 50 percent above the median, or zero link clicks on more than 250 HUF of spend.

Copy

• Keep when the text sits on ads at or below the median CPC.
• Alter when the hook repeats a text already in rotation, or the body runs over 700 characters.
• Cut when the text only rides ads that are being changed, or duplicates a better text. Never cut on per-body numbers, Meta does not report them.

Audiences

• Double down at or below the median cost per lead with at least 3 leads, raise budget by at most 30 percent per week.
• Keep within 50 percent of the median.
• Exclude more than 50 percent above the median on at least 30 clicks, or when the interest set only cannibalises a winning ad set.

Thresholds this month

• Median CPC 130.18 HUF, median cost per lead 11 635 HUF, both august account level.
• Minimum evidence 30 link clicks per arm, or 3 leads once the Lead event is live.
• Below that the answer is always keep running, never cut.

1

Browser side, verified on the local staging copy. ViewContent per product category from the URL slug, Contact on mailto clicks, matching view_item and contact_click dataLayer pushes, and form actions rewritten at runtime so a cached page cannot pass one visitor's click ID to the next.

2

Lead with a shared event id. The id lead_email_date is stamped onto the Lead the existing GTM container already fires, so the Conversions API copy deduplicates instead of double counting. PageView, telefon Contact and a second Lead are deliberately not re-fired.

3

Server side, read back from Meta. Conversions API from the Apps Script for Lead, QuoteSent and Purchase, all events_received 1 under test event code TEST_FUSION_E6, fbtrace ids in qa/measurement/capi-readback.txt. Google Ads values moved from 1 HUF to 0 on 6 actions, two offline actions created for the quote and order path.

4

Score. 4 of the 5 audited defects fixed, the fifth partial. Snippet 3101 is staged with the two command deploy path quoted below and goes live on deploy. Until then every cost per lead outside Meta's own lead surfaces stays not available.

Audiences

Nothing is excluded and nothing is doubled down on this month, because one ad set carried 100 percent of august spend and there is no contrast to act on. Split the 20 interest set into two ad sets for september, billiard and cue sports against interior design and luxury furniture, so october has a real comparison. Keep age 25 to 65, Hungary, all genders.

Google

PMax search category impressions fall from 1 791 in july to 47 in august while cost nearly doubles, and july's categories were real intent terms while august is mostly the unlabelled bucket plus the brand term. Recommend re-enabling the paused 2605-Fusion-Tables_Search campaign on biliárd asztal, biliárd asztal ár and használt biliárd asztal.

30 napos eredmények

• The 316 and 976 index values and the work counters are carried over from the accepted v2 report, not recomputed from raw platform logs.

Hirdetéskreatívok teljesítménye

• The CTR ranking excludes four ads below the stated minimum impression floor.
• The duplicate-upload pattern is stated as a correlation, not as a cause.

Versenytársak hirdetései

• biliardcentrum.hu Meta presence is not covered by the snapshot, so it is labelled not available rather than zero.
• 4Biliard.hu individual Google ad start dates are not in the snapshot, only the aggregate count of four.

Hirdetési és célzási stratégia

• The monthly baseline B is unverified because the Meta API token is dead, so every allocation is a percentage with one historical illustration.
• Interest targeting categories are proposals, no current audience size measurement was possible.
• Five non-billiard landing pages are proposed, not confirmed live.

Persona elemzés

• Persona 4 has no age or gender anchor, it is anchored on region and platform only.
• Budapest and Pest combined clicks and form starts are not available, only the verified spend share.

Belső anyagok és QA

• The five-ranges finding id was read from the audit's top-10 table, not from the full consolidated findings file.