Rule bot_class_gating, from the product code

Bot traffic beyond the known-bot filter

Googlebot is filtered before counting starts. This rule catches the rest: scrapers, headless browsers and click farms that mimic people well enough to count.

Rule 7th of 12Holds, never deletesEvidence in your queue

Fraud queue on demo data: bot_class_gating sits with its verdict, subject, reason and action, above the held commissions with amounts.

The fraud queue, running on demo data.

How the rule works

The exact mechanics, from the engine. No black box, no tuning required.

  1. 01After the known filter

    Verified crawlers are excluded up front. What's left gets scored on behaviour signals in the user agent and click pattern.

  2. 0250% share, 20-click floor

    Half of a meaningful sample must look automated. Small affiliates with odd traffic never trip it.

  3. 03Money follows the clicks

    Commissions traced to the flagged clicks are held with the account, so bought traffic can't convert its way to a payout.

The signals

  • 50%+ suspected-bot share
  • 20+ counted clicks in 7 days
Holds, never deletesA hold freezes the commission before payout. Nothing is lost while you decide.
Evidence attachedThe queue shows exactly what tripped, side by side with the sale.

What lands in your queue

When bot traffic fires, this is what you see. Release or uphold is your call, not the engine's.

Fraud queue · bot_class_gating · demo dataMade-up numbers
Example held commissions after bot traffic fired, on demo data
AffiliateWhat trippedCommissionStatus
Affiliate 2731 of 51 counted clicks scored as bots$76.50Held
heldWaits for youThe commission stays out of payouts until you decide.
releasedPaid normallyReal pattern after all. It returns to the pipeline.
upheldNot paidFraud confirmed. The clawback is journaled.

In the wild

Bots are a third of everything

  • Stat: Imperva 2024: bad bots were about one third of all internet traffic in 2023. Source
  • Case: United States v. Michael Smith (pled guilty Mar 2026): a botnet streamed AI-made songs billions of times for ~$10M in royalties. First streaming-fraud conviction.

Cases and figures verified against court records and named reports, . Vendor estimates are labelled as such.

How attackers try to beat it

Headless Chrome sends a perfectly normal user agent. The classification also weighs patterns a script can't easily fake consistently across thousands of clicks.

Every rule is one layer. The twelve overlap on purpose: beating one cleanly usually means walking into another.

Related rules

Rule thresholds checked against the code . Windows are rolling.

Bot traffic questions

Can I change the threshold?

The thresholds ship fixed, tuned from the rule code and tested together. Tightening one in isolation breaks the overlap the twelve rules build as a set.

What does a hold cost me?

Nothing but time. A held commission is frozen, not deleted. If you release it, it flows to the next payout exactly as if the hold never happened.

Does this rule fire on my best affiliate?

Every rule is scoped to the commissions linked to what tripped it, weighted by conversions, and rate-limited on the click side. A rival can't aim it at your top earner.

This rule, on every sale,
before the payout.

Self-host for $99 once, or run it in the cloud free for 14 days.