A post hoc selection pressure on a true signal generates an untrue result.

For example, man on the street style interviews by a biased producer.

"But it's real people saying that!" / "Yes, and they could be a tiny minority of the population "

With enough data, and enough filtering, any signal can say anything else.

You incorrectly think, "but it is derived from a true signal", but the heavy-handed selection dominates the underlying signal.

This also happens when looking at data on internal dashboards.

It's very easy to accidentally lie to yourself based on the construction of your metric.

"I'm just doing what the data tells me to" is a comforting fiction that implies that the synthesis and distillation of signals does not include many implicit subjective decisions.