Traffic quality and trust

What we monitor, and what it never does

We watch the traffic behind every athlete's tracking link so the sales you pay for are real. Five signals run continuously. All five run in shadow mode: they record and surface what they see, and they never reverse a commission, block an athlete, or change your invoice on their own. A person makes every money decision.

This page covers the five signals, why they run in shadow mode, the nightly anomaly check, and what happens when an admin confirms fraud.

The five signals we watch

Each signal looks at the traffic on a tracking link from a different angle. Together they describe traffic health; alone, none of them is a verdict.

Velocity

Velocity measures how fast clicks arrive from a single source. Real audiences click over hours and days. A wall of clicks in a few seconds from one place stands out. We flag the burst; we do not act on it.

IP correlation

IP correlation checks whether a cluster of clicks shares one IP address. A real audience spreads across many networks. Many clicks from one address is a sign that one machine is doing the work of a crowd.

Fingerprint reuse

A device fingerprint is a rough signature of the browser and device behind a click. We store it as a one-way hash, not anything that identifies a person. When the same fingerprint shows up over and over, it suggests one device is posing as many visitors.

Bot score

The bot score rates how likely a click came from automation rather than a person. The Cloudflare Worker that handles the redirect computes it at click time and stores it on the click record, so the signal is captured the moment the click happens.

Click quality

The click quality score rolls the others up into one sortable number per athlete. It weighs IP and fingerprint diversity, bot health, and whether conversions track sensibly with clicks over time. It is a signal you sort by, never a gate and never an automatic clawback. For how the click quality score is defined and used, see verified and quality-scored partners.

Why shadow mode

A money product should not reverse a payout on a guess. So the fraud signals run in shadow mode: they log what they see and surface it for review, and they never gate a payout or claw back a commission by themselves.

Shadow mode protects both sides of a match:

• For you, it means flags are evidence you can act on, not automatic charges you have to dispute after the fact.
• For the athlete, it means an unusual but honest week never silently costs them a payout.
• For the ledger, it means every reversal has a reviewed decision and an audit trail behind it.

Nightly anomaly detection

Once a day, an automated check compares each athlete against their own recent baseline, not against a fixed platform-wide threshold. That matters: a small athlete and a large one have very different normal volumes, and a single rule would punish one and miss the other.

The check uses a robust median method (median plus median absolute deviation, a modified z-score) instead of a simple average. A plain average is easy to skew with one big day; the median method tolerates normal spikes and only flags traffic that is genuinely out of pattern for that specific athlete.

When the check finds an anomaly, it records a fraud signal for an admin to review. It does not reverse anything. The output is a flag, never a charge.

What happens to confirmed fraud

Confirmation is a human decision. An admin reviews the flagged signals in the fraud queue and either dismisses them or confirms fraud. Nothing happens automatically before that.

When an admin confirms fraud, it routes into the same dispute and on-ledger writeoff path you already use for refunds:

• The commission is reversed on the ledger, with an audit record of who confirmed it and why.
• It flows through the existing dispute and writeoff path, the same mechanism behind a refund, so the accounting is consistent and traceable.
• There is no silent auto-clawback. A payout or invoice line only changes because a person decided it should.

This is the same reversal machinery as a refund, just triggered by a confirmed-fraud decision instead of a Shopify refund. For how that ledger reversal works end to end, see how refunds and clawbacks work.

FAQ

How do you stop fraud?

We watch five signals on the traffic behind each athlete's tracking link: click velocity, IP correlation, fingerprint reuse, bot score, and a click quality score. The signals run in shadow mode, so they record what they see and never reverse a commission on their own. A nightly anomaly check compares each athlete against their own baseline using a robust median method, so one busy day does not trigger a false flag. Anything that looks off goes to a human admin to review.

Will real athletes ever be penalized by mistake?

No automatic penalty is possible. Every fraud signal is shadow mode only: it logs and surfaces, it never reverses a payout or blocks an athlete. The nightly anomaly check uses a median-and-MAD baseline that tolerates normal spikes, and the click quality score is a sortable signal, never a gate or an automatic clawback. A real athlete with an unusual but legitimate week is reviewed by a person, not punished by a script.

What happens to confirmed fraud?

Only a human admin can confirm fraud. When they do, it routes into the same dispute and on-ledger writeoff path you already use for refunds, not a silent automatic clawback. The commission is reversed on the ledger with an audit trail. Nothing leaves your invoice or an athlete's payout without that reviewed decision behind it.