# Reconciled diagnosis

The evidence establishes a delivery interruption, followed by restored traffic without a later successful paid consultation in the inspected source cohort. The current paid booking itself is now proved, and its exact source identifier subsequently arrived once in raw row177 through existing automatic processing. The evidence does not isolate the cause of the historical post-restoration payment gap.

## Comparable funnel

| Budapest period | Days | Spend HUF | Impressions | Outbound clicks | Landing-page views | GA4 Meta sessions | CTA events | All-channel paid consultations / matched rows |
|---|---:|---:|---:|---:|---:|---:|---:|---:|
| August 1–25 | 25 | 50,444 | 106,847 | 1,403 | 1,247 | 1,427 | 60 | 9 / 9 |
| August 26–September 20 | 26 | 509 | 978 | 14 | 13 | 17 | 1 | 0 / 0 |
| September 21–25 | 5 | 13,825 | 12,800 | 257 | 209 | 235 | 9 | 0 / 0 |
| September 26–27 | 2 | 9,047 | 9,299 | 171 | 150 | 117* | 12* | 0 / 0 |

*The last GA4 period includes known QA contamination and is not evidence of improved genuine intent. September26 is also a partial change day. September28 is excluded from comparisons. Payment-source coverage starts August1 at02:00 Budapest, leaving its first two hours unavailable, and ends September28 08:52:54 Budapest. Payment counts are source-created cohorts at current paid/captured status. They are not Meta-attributed sales, all inquiries, attendance, or a complete Calendar booking history.

## Confirmed and contributing evidence

- **Account-level delivery interruption:** exact Meta account history changed Active→Payment Needed on August26, Payment Needed→Disabled on August30, and back to Active on September21. Delivery collapsed over that interval. Billing charge and chargeback entries exist, but the prior exact-account unusual-activity/no-payment-due finding prevents confidently labelling the underlying cause an unpaid balance. A charge event alone is not settlement proof.
- **Traffic returned before Friday's complaint:** September21–25 has257 outbound clicks,235 paid-source sessions and9 booking-CTA events. It has zero later successful paid consultations in the inspected all-channel cohort. September26 creative and website edits therefore cannot explain a complaint already present September25.
- **Acquisition became more expensive:** outbound CPC increased from35.95 to53.79HUF across the productive and restored periods, and CPM from472.11 to1,080.08HUF. Paid-source sessions per day decreased from57.08 to47.00. These are contributing delivery/efficiency observations, not an isolated auction or creative cause.
- **No gross engagement collapse shown:** paid-source engagement is57.8% before interruption versus60.0% after restoration. CTA-events/session is4.20% versus3.83%. The samples are unmatched and small, and events are not unique people. They cannot establish a statistically reliable conversion deterioration.
- **The missing-Sheet-only explanation is not supported for paid consultations:** nine successful August payments match nine distinct rows. No later successful payment lacking a row was found. A historic August27 generic sync error does not establish a continuing outage.
- **Offer and current availability:** the5,000HUF offer existed in August paid checkouts, before our edits. The live Calendar page currently exposes available slots and the real QA form reached the correct merchant's5,000HUF checkout. Historical availability and user abandonment motives remain unknown.
- **Current tracking repair remains in production:** exact published GTM resource and repaired CTA rule match verified prior tests. A fresh click reaches GA4 once and the exact Meta browser receipt once. This repairs intent measurement, not proof of paid-booking recovery.

## Unresolved causal and operational limits

The evidenced gap is after recorded CTA intent and before successful paid consultation. There is no identity-level join between the nine restored-period CTA events and actual customers, so it is not a nine-person abandonment finding. The available history cannot separate ordinary small-sample variation, booking-page friction, offer fit or historical availability. The real QA payment and booking completed at09:43 Budapest on September28. Google shows payment and booking confirmed, Stripe reports succeeded/captured5,000HUF, and both provider emails arrived. The existing authorization required the normal return in the original Google session. This test recovery does not prove historical customer abandonment had the same cause. Two early raw reads were still empty through07:53:36UTC. At08:06UTC the full bounded source-ID scan found exactly one match, and row177 contains the correct QA identity, contact, service, paid marker, appointment time and Calendar UID. No manual append, notifier substitution or sync repair occurred. This is observed delayed normal intake, with the configured cadence unknown. It rules out a complete current failure of this tested route, but cannot establish historical availability, every customer journey or why the restored-period genuine payments were zero. One historical slot mismatch and three payer/attendee name differences are retained as unresolved reconciliation details, not silently called exact-field matches.

D1.1 is satisfied by the reproducible dated funnel and explicit missing-data definitions. D1.2 remains PARTIAL because the post-restoration causal diagnosis remains insufficient. No future lead-volume guarantee is made.

Private supporting reports: ads-report.md, source-report.md, tracking-report.md. Reproduction: evidence/ads/normalize_funnel.py and evidence/source/reconcile_saved_sources.py. These operate on saved exports and perform no provider writes.