Reading Your Traffic-Quality and Click-Quality Signals | Harmonia
How Harmonia monitors affiliate traffic quality: velocity, IP correlation, fingerprint reuse, bot score, and click quality. Signals inform your decision and never auto-reverse a commission.
Harmonia watches five signals on the traffic each athlete sends: click velocity, IP correlation, fingerprint reuse, a per-click bot score, and a per-athlete click quality score. A nightly check also compares each athlete against their own baseline. All of it runs in shadow mode, so a flag surfaces to admins and to you without reversing a commission or blocking an athlete on its own. When a real problem is confirmed by a person, it routes through the normal dispute and write-off path, not an automatic clawback. This guide explains how to read each signal and what to do when one trips.
What the platform monitors
Every click and order an athlete sends is checked against a few patterns that separate real shoppers from automated or duplicated traffic. None of these patterns moves money by itself. They feed the click quality score and an internal review queue.
| Signal | What it looks for | What a flag suggests |
|---|---|---|
| Click velocity | Bursts of clicks too fast or too regular to be human | A script or bot, not a person |
| IP correlation | Clicks and orders sharing one network beyond what real shoppers do | Traffic from a single source dressed up as many |
| Fingerprint reuse | One device fingerprint across clicks that should be different people | The same device posing as a crowd |
| Bot score | A per-click read of how automated each click looks | The individual click may be machine traffic |
| Click quality score | A per-athlete roll-up of the signals above | Lower-quality traffic overall, worth a closer look |
You can sort applicants and partners on the click quality score. The other signals live in the review queue and the traffic quality and trust panel, where a flagged athlete shows the pattern that tripped it.
Click velocity
Real people click at human pace and spread out over time. The platform flags bursts that are too fast or too even to be a person, the shape of a script firing on a loop. See the click glossary entry for what counts as one.
IP correlation
Real customers come from many networks. When a cluster of clicks and orders shares one network in a way genuine shoppers usually do not, the platform flags it. This catches traffic from a single source made to look like many separate people.
Fingerprint reuse
A device fingerprint is a rough signature of the browser and device behind a click. Different people carry different fingerprints. When the same fingerprint shows up across clicks that should be separate shoppers, the platform flags the reuse.
Bot score
Each click gets a bot score, a read of how automated it looks on its own. A high bot score on one click is a data point, not a verdict. It feeds the click quality score rather than standing alone.
Click quality score
The click quality score rolls the signals above into one per-athlete number, built from patterns like IP and device diversity, bot signals, and whether conversions look like real shoppers. It is a sortable signal you use to compare partners. It is never a gate and never a clawback. For how it sits alongside identity verification and the LLM quality score, see verification and quality scoring explained.
Shadow mode: signals inform, they never move money
Every signal above runs in shadow mode. The platform monitors traffic, scores it, and surfaces flags to admins and to you. It does not reverse a commission or block an athlete on a signal alone.
- Nightly anomaly detection compares each athlete against their own baseline and flags a sharp shift, the kind of change worth a human look.
- A flag opens a review, not a penalty. It lands in an internal queue for a person to judge.
- No score gates a payout. Browsing, applying, and earning are not blocked by a quality signal. Only the dispute and write-off path can reverse a commission, and only after a person confirms a real problem.
Signals inform, you decide
Velocity, IP correlation, fingerprint reuse, the bot score, and the click quality score all help you choose. The platform never auto-approves, auto-declines, or claws back a commission on a score alone.
What happens when fraud is confirmed
A flag is a question, not an answer. A person reviews the pattern before anything moves. Once a case is confirmed, it routes through the standard dispute and write-off path, the same path a refund uses, not an automatic clawback.
That ordering matters for two reasons:
- A real athlete is not penalized for one odd-looking day. A single burst or a shared network does not reverse earnings. It opens a review, and a person decides.
- Confirmed cases follow one known path. Disputes and write-offs are the only routes that adjust a commission for a quality reason, and a person triggers them. For how reversals work, including refunds inside the refund window, see refund handling.
The only thing that reverses a commission automatically is a refund inside the program's refund window, pro-rated to the amount refunded. That is a money event, separate from any quality signal.
How this fits a curated program
These signals are the running layer of a curated program. You choose your partners up front through the two-way veto, and you keep reading the traffic they send once they are live. Because attribution is first-party and server-side, with no discount code for the customer to enter, the order data behind these signals is your own, not a third party's guess.
When you read a flag, weigh it against the rest of the partnership: the athlete's history, their fit with your category, and whether the pattern repeats. One signal is a prompt to look closer, not a reason to act alone.
What you pay, and what a flag does not change
A flag never changes what you owe. You pay the commission you set plus a 20% platform fee on top, billed to you and never deducted from the athlete, so the athlete receives the full commission. The fee is charged only when an athlete drives a real attributed sale. A quality signal does not add a charge, remove one, or reverse a payout on its own.
For the full breakdown of the commission and the fee, see what brands pay: the commission pool and the fee.
FAQ
How does Harmonia detect affiliate fraud?
The platform watches five signals on every athlete's traffic: click velocity, IP correlation, fingerprint reuse, a per-click bot score, and a per-athlete click quality score. A nightly anomaly check also compares each athlete against their own baseline. All of this runs in shadow mode, so a flag surfaces to admins and to you without reversing any commission or blocking any athlete on its own.
Will a real athlete ever be penalized by mistake?
No. Signals never move money on their own. A flag opens an internal review, not an automatic clawback or block. A real problem is confirmed by a person before anything happens, and confirmed cases route through the normal dispute and write-off path. An athlete with one odd-looking day is not penalized for it.
What happens when fraud is confirmed?
It routes through the standard dispute and write-off path, the same path a refund uses, not an automatic clawback. A person confirms the case first. Nothing is reversed on a score alone.
What does the click quality score measure?
It summarizes the quality of the traffic an athlete sends, built from patterns like IP and device diversity, bot signals, and whether conversions look like real shoppers. It is a per-athlete sortable signal you use to compare partners. It never blocks an athlete and never triggers an automatic clawback.
Do quality signals automatically reverse commissions?
No. Every quality and fraud signal runs in shadow mode: it informs your decision and an internal review queue, and it never reverses a commission or pays one back on its own. The only thing that reverses a commission automatically is a refund inside the program's refund window, pro-rated to the amount refunded.