DUPLICATE CONVERSION REVIEW
Working checklist, not a Google upload format or account-changing script.

FREEZE THE COMPARISON
Account / campaign / actual selected standard or custom goal:
Date range / account and CRM time zones / review cutoff:
Conversions versus All conversions versus conversion-time reporting:
Attribution model / windows / business outcome / value definition:
Order source of truth and any test/refunded/cancelled order treatment:

MAP EACH PATH FOR THE SAME OUTCOME
Business order ID (opaque; no email or other customer identity):
Event trigger / original source / consent state:
Target account and conversion action identifier / owner:
Browser tag / GA4 import / platform plugin / server or Data Manager path:
Exact transaction_id sent on each path (missing stays missing):
Value / currency / units (dollars and cents are different):
Primary/secondary setting / campaign goal / custom-goal membership:
Count setting / actual destination diagnostics / evidence timestamp:

CONTROLLED VALIDATION
One confirmed order: correct ID, value, currency and action:
Reload/revisit same confirmation: original business ID stays stable:
Second distinct order: new business ID, even from the same ad click:
Different sources: same intended action and exact ID where supported:
Separate action or wrong ID: investigate, do not assume a shared merge:
Missing/empty/static/placeholder ID: preserve evidence and fix mapping:
GA4 purchase path: web versus app stream and collected user context:
Test activity treatment and validation environment documented:

CHANGE RECORD
Chosen authoritative purchase outcome and delivery owner:
Reason for retaining/removing an overlapping path from bidding:
Campaign-specific and custom goals reviewed before changing settings:
Tag/data-layer change / exact scope / owner / rollback:
If multi-source: feature eligibility, destination action, consent fields,
  exact ID match, value/currency/units and trial behavior verified:
Change date; before/after reports; reporting lag and business reconciliation:
Do not infer historical data is repaired merely because a setting changed.
Do not add an extra sender to compensate for an untraced missing event.