My toaster has a slider.
It controls how dark the toast gets.
The toast does not get a vote.
Fraud platforms have a slider too. It controls how aggressive the detection is. The clicks do not get a vote either.
You drag it left to be "more aggressive." You drag it right to be "more conservative." The vendor may call it a confidence threshold, a sensitivity setting, or something even more scientific.
It is extremely satisfying to move.
A slider is not a finding.
A click either came from a browser lying about what it is, or it did not. That is a fact about the world. A slider does not change the fact. It changes how much of the vendor's uncertainty you are willing to pay for.
When you buy traffic, you are not buying a probability distribution. You are buying a sequence of events: a user saw something, clicked, and either was a real person or was not. You need to know which.
A confidence threshold does not answer that. It answers a different question:
How much doubt are you comfortable with?
That is a question about you. Not about the click.
The vendor's model produces a number. The slider decides which numbers get acted on. But the number itself is not evidence. It is a summary of the model's internal state.
"Invalid, 73% confidence" is not more transparent than "invalid." It is less.
Because you have now been handed a number you cannot audit, cannot challenge, and cannot take to the partner who sold you the traffic.
You can, however, be held responsible for it.
That is the structural consequence.
Once the threshold is your setting, every argument about a wrong call becomes an argument about your configuration.
Blocked a real customer? You set the slider too aggressively.
Let a bot through? You set it too conservatively.
The vendor did not make the call. You did.
No one has to intend this.
A slider is a perfectly reasonable response to a buyer who asks for control. It is also a perfectly reasonable response to a model that outputs a score. But the structure has a consequence.
It moves the argument from "what did you find" to "what did you set."
That is a much less useful argument.
Mathematically, a threshold is just a trade-off. Lower it and you block more, catching more fraud and more humans. Raise it and you block less, letting more fraud through and more humans through. That trade-off is real. But it is not a detection result. It is a budget.
And a budget is a decision about risk. It is not a fact about a click.
The real question in invalid-traffic detection is not how aggressive you want the system to be. It is whether the system can tell you what it observed and why that observation means the click was invalid.
A threshold is a dial on a black box.
A dial on a black box is not control. It is a mood ring with numbers.
You can turn it all day. The box will not tell you what it saw.
And that is the information you actually need when you go back to the partner who sold you the traffic.
You cannot have a conversation about a probability. You cannot negotiate with a confidence score. You cannot put a 73% on the table and ask why it is not 74%.
You need a fact.
This browser reported itself as Safari on iOS, then answered a question only Chromium can answer.
That is something you can take to the counterparty.
That is something you can have an argument about.
That is something that either happened or did not.
A slider does not produce any of that.
It produces a setting.
And the setting is always yours.
So here is a small experiment for the next time someone shows you a threshold.
Ask what happens when you set it to the middle.
Then ask what happens when you set it to the left.
If the answer is the same sentence with different numbers, you have not been given a detection system.
You have been given a dimmer switch.
And the light still has not been turned on.