Traffic source provenance: checking the supplier chain
Check promised placements and intermediaries with a source record, an evidence review and two full-cost examples with reversals and stop criteria.
A placement sold as an editorial recommendation may arrive through an agency, a subpublisher and a redirect service. The advertiser needs to know whether that route matches the agreement. A strange domain in a report is a reason to reconcile evidence, not a verdict of fraud. This guide builds a source record, a fair investigation procedure and a capped pilot with two complete financial examples.
The educational setting is a German business software service receiving voluntary trial registrations from professional publications. It is not a gambling acquisition guide. The workflow can inform other markets, but source permission, contracts and data requirements need local verification. USD amounts are independent teaching assumptions for this English edition, not German market prices or currency conversions. Every case, threshold and template below is an editorial construction, not an AffiliateCore customer result.
1. Define exactly what was purchased
Separate the commercial supplier, the owner of the audience, the actual placement and the tracking intermediary. They can be four different entities. An agency can legitimately invoice you for a publisher's article; that does not make the agency website the original traffic source. Conversely, a recognised tracking domain does not establish that the promised publisher carried the campaign.
Write an observable promise: one identified article on an approved publication, one visible commercial disclosure, an agreed destination and a defined publication period. Specify whether related pages, newsletters, syndicated copies or additional subcontractors are included. A promise of “premium European traffic” cannot be reconciled because it never defines inventory. Name the publication and format rather than accepting a supplier's broad quality label.
Our AffiliateCore platform can provide the advertiser offer, approved campaign context and conversion records for the supported process. It does not replace evidence of where a reader saw a placement. Before setting up the offer, prepare the commercial description and the source acceptance criteria. If no suitable product or permitted source is available, complete that preparation without promising a live campaign.
2. Create the source record before launch
Use one record per placement, not one vague description for an entire network. Suggested fields are supplier reference, audience owner, publication URL, placement code, creative version, destination, country, start and end dates, permitted intermediaries, responsible contact and evidence location. Record the agreed billable event and the review deadline separately. A source record should describe an agreement that someone can actually check.
For the teaching pilot, the brief says: publication P17, article slot M04, creative C02, German business audience, voluntary trial registration, no additional publication without written approval. The supporting material is an article preview explaining a software workflow, followed by a clear invitation to request a trial. It does not promise guaranteed savings or disguise the commercial relationship. The final URL and actual published wording must match the approved version.
Maintain a dated change history. An approved new placement gets a new identifier; do not recycle the old code merely to keep reporting convenient. Give the supplier a way to request a change before sending traffic. This avoids turning a normal inventory adjustment into an avoidable dispute. A spreadsheet is sufficient for a small pilot if access is controlled and every decision has an owner.
3. Know what each piece of evidence can establish
MDN documents that Referrer-Policy can omit the referrer or restrict it to the origin. Consequently, an empty referrer is inconclusive. Do not ask a partner to weaken browser privacy merely to make your report look complete. Compare the placement record, authorised publication evidence and corresponding time window instead.
A sub-ID is an agreed label, not independent certification. A screenshot records what somebody captured at a particular moment; by itself it does not prove the entire delivery period. An invoice demonstrates a commercial claim, not every visitor's route. A conversion identifier connects records, but does not establish that a visitor read a particular article. Treat these as complementary observations with different limitations.
Create three evidence columns: “supplier declared”, “we observed” and “still unexplained”. Never overwrite a declaration with your inference. Keep original files, their dates and the exact question under review. For a supplier dispute, a compact sample with relevant identifiers is usually more useful than a large export of customer records. Share only the information needed for the agreed investigation.
4. Review intermediary chains without confusing standards
Awin's subnetwork code requires subpublisher identification through clickref, source URL disclosure and responsibility for subpublishers. Those are Awin-specific obligations, not proof of an AffiliateCore integration. For your own agreement, identify the contracting party and each approved audience owner, then decide how additions will be reviewed. Do not assume that joining a network authorises every downstream supplier.
In programmatic inventory, ads.txt declares authorised sellers. IAB Tech Lab describes sellers.json and SupplyChain as tools for identifying selling participants. These sources support checking the relevant advertising supply path; they do not certify an affiliate registration. Apply them only where the bought inventory actually uses that ecosystem. A direct editorial placement without an auction does not become fraudulent merely because no bid-chain record exists.
Ask for evidence appropriate to the transaction: publication confirmation for an article, the relevant account and inventory identifiers for programmatic buying, and a disclosed supplier chain for a delegated placement. When information cannot be shared publicly, agree a restricted review route. “Confidential” may justify access controls; it does not make an unverifiable placement automatically acceptable.
5. Run a controlled reconciliation
Start with a seven-day pilot, one approved placement and a written spending ceiling. These durations are recommendations, not platform requirements. Check a permitted preview before launch and agree how technical checks are excluded from payable results. Do not manufacture registrations, self-referrals or commissions to test a commercial offer.
During the pilot, compare daily totals across supplier delivery, your landing analytics and accepted business events. Use one time zone and document the cut-off. A click total need not equal session totals: define what each system counts before alleging a discrepancy. Compare like-for-like periods, remove documented technical tests, and distinguish duplicate delivery notifications from duplicate customer actions.
In AffiliateCore the supported click flow accepts three sub-ID fields. A practical convention is source, placement and creative. Keep the mapping outside the public URL and never put email addresses or names into those labels. The labels help group records; they do not authenticate the publisher. If you are ready to prepare an offer, register as an advertiser with the source record and acceptance conditions already written.
Before launch, reserve USD1,510 for case one and USD2,264 for case two: these are the respective fixed costs plus provisional variable liabilities before reversals. Treat them as maximum authorised exposure for the stated cohorts, not targets to spend. Stop new delivery before a forecast exceeds the cap. Within these teaching assumptions, the listed amounts include applicable non-recoverable acquisition taxes. No additional payment, foreign-exchange or other charge is assumed; add any actual expense not already included. This is not a claim that real supplier prices universally include tax. Recoverable VAT and payment timing belong in a separate cash plan. Do not fund obligations on an assumed future reimbursement.
6. Example one: an unusual domain with a legitimate explanation
Assume the German software service approved one professional publication through an agency. Its report shows an unfamiliar redirect domain. The reviewer pauses any expansion, preserves the observation and asks the agency which approved component uses that domain. The publication supplies its dated placement confirmation; the agency provides the corresponding route description and matching placement code. The evidence resolves the route discrepancy without identifying it as fraud.
The teaching cohort contains 120 recorded registrations. Twelve are duplicate records and eight fail the previously agreed qualification rules. That leaves 100 preliminarily accepted registrations. Ten are later reversed for documented ineligibility within the agreed review process, leaving 90 final accepted registrations. No cases remain pending. The duplicate and rejected records are not paid, and the ten reversals reduce the provisional commission rather than being charged twice.
Commission is an assumed USD10 per final accepted registration, and the separate platform/service charge is an assumed 10% of that commission. Neither is a quoted AffiliateCore tariff. Full cost comprises USD900 commission, USD90 service charge, USD150 placement, USD120 setup work (six hours at USD20), USD80 review work (four hours at USD20), USD30 allocated tools and USD30 reversal handling. Total cost is USD1,400.
The provisional 100-registration variable charge was USD1,100. Reversing ten registrations removes USD110, giving the final USD990 variable charge included above. This demonstrates the adjustment explicitly; reversal handling remains a real expense. Cost per final accepted registration is USD1,400 / 90 = USD15.56, rounded to cents. Against a preset USD17 ceiling, the cost gate passes.
Assume each final accepted registration contributes USD22 after expected service delivery, payment costs and allocated non-acquisition variable expenses. This is a teaching valuation, not booked sales revenue. Ninety times USD22 is USD1,980; subtracting acquisition and review costs leaves USD580 before general overhead and profit taxes. The next decision is another pilot at the same size, not immediate scaling. One resolved domain discrepancy does not establish future delivery quality or incremental demand.
7. Example two: undisclosed placement and a failed cost gate
A second agency supplies the same type of German trial registration, but the agreed placement is a particular professional article. A review finds that one source label covers two publications. The supplier confirms that 40 of the final registrations came from a second, unapproved publication. Preserve the admission and supporting delivery records. This establishes a scope breach for those placements; it does not by itself establish fictitious people or malicious intent.
The cohort starts with 150 records. Twenty duplicates and ten qualification failures leave 120 preliminarily accepted registrations. Twenty further registrations are reversed under the original eligibility rules, leaving 100 final valid registrations. Sixty belong to the approved publication and 40 to the undisclosed one. For this educational contract, valid registrations remain payable while the placement breach is handled separately. Do not retrospectively invent a rejection rule just because the supplier relationship deteriorated.
Final variable cost is 100 × USD12 commission = USD1,200, plus 10% service charge = USD120. Add USD300 placement, USD200 investigation labour (ten hours at USD20), USD80 tools and USD100 correction/reversal handling. Full cost is USD2,000, or USD20 per final valid registration. The prior 120-record variable liability was USD1,584; twenty reversals remove USD264 and leave USD1,320. No reimbursement for the placement breach is assumed or netted off before it is agreed and received.
With an assumed contribution of USD18 for each valid registration, the cohort contributes USD1,800 before acquisition, leaving a USD200 loss. The preset cost ceiling is USD18, so economics fail independently of the source breach. Stop new supply from the disputed route, honour valid liabilities and request a documented corrective plan. Any commercial refund is recorded separately if it occurs.
A useful follow-up test would isolate the approved publication, use a fresh placement code and include all review costs. It is not justified merely by the claim that the cheaper undisclosed publication brought “more volume”. Scaling the mixed cohort would hide both the source problem and the unprofitable unit cost. The two issues need separate evidence and separate decisions.
8. Make the dispute procedure fair and reproducible
Open a case with the placement identifier, date range, contractual promise, observed discrepancy and requested evidence. A neutral request might read: “Placement M04 was agreed for publication P17. The delivery record also references P23. Please explain the route and provide the relevant placement confirmation.” This asks an answerable question without accusing the supplier of fraud.
Set a response deadline suited to the risk and contract; a proposed two-business-day review is an internal operating choice, not a universal legal standard. Pause new disputed inventory if needed, but keep undisputed liabilities separate. If an explanation arrives, record what it resolves and what remains uncertain. Have a second reviewer check any financially significant conclusion.
Use outcomes such as explained, delivery error, contractual breach, unresolved or demonstrated deception. Do not collapse them into a single “bad traffic” label. A wrong code can be corrected with a recorded mapping; it should not be silently rewritten across historical reports. If evidence remains insufficient, limit future exposure and escalate through the agreement rather than publishing allegations.
9. Keep business decisions separate from technical status
AffiliateCore records supported conversions and their status; an incoming callback is not automatic proof of an approved source. The external API is read-only and is not a complete source-evidence or financial settlement feed. The source record and documentary review therefore remain a separate operating process. Do not promise an automatic supplier-provenance investigation button.
Likewise, a refund event string is not a supported cashback lifecycle simply because an endpoint accepts text. For this guide, the legitimate registration event and supported review workflow are the relevant starting point. Confirm the actual offer configuration before launch. If two systems disagree about a later reversal, reconcile the known event rather than sending a new event and hoping that it cancels the original.
Existing advertisers can sign in to AffiliateCore to review their offer and campaign context, then attach the relevant evidence through the agreed support process. Keep customer data out of public screenshots. The aim is a decision that another reviewer can reproduce, not a dashboard that removes every uncertainty.
10. Decide what to repeat and what to stop
Before increasing spend, require a resolved source map, final cohort numbers, full costs below the agreed ceiling and a manageable review workload. Test one change at a time. An unexplained route change, an unwillingness to identify the placement or a failed financial gate stops expansion even if registration volume looks attractive. A low cost is not a substitute for permission.
Your practical assignment is to complete one source record and trace five authorised sample records from placement to final decision. Note missing evidence without collecting unnecessary personal information. Recalculate both worked examples with your own agreed rates and actual reversals. Then write a short decision: continue unchanged, retest a corrected route, or close the placement. State which evidence could change that decision and who is responsible for obtaining it.
Do not claim additional customers from attribution alone. The examples evaluate attributable cohort economics; incrementality requires a separate, properly designed comparison. Close the pilot only after late reversals and unresolved liabilities have been addressed under its terms. Preserve the original report so a later correction remains visible rather than erasing the history.
Sources checked 18 October 2026.