For most of television history, nobody actually counted the audience. A panel of a few thousand households was measured, and the rest of the country was extrapolated from them. That is a reasonable trick when direct counting is impossible, but every panel is a sample, and every sample carries the bias of who got picked.

Those biases were not random. Panel-based ratings skewed by class, race, age, and education, because who ends up in a measurement panel is never a neutral draw. So networks told everyone (advertisers, investors, and crucially themselves) a distorted story about who was watching. The map was mistaken for the territory for decades.

Server math does not flatter you

Streaming changed the instrument. A stream is a logged event on a server, not an inference from a panel. You are no longer estimating the audience from a sample; you are counting it. That number is truer, and it is also scarier, because it strips away the comfortable extrapolations the old system let networks hide behind.

A panel tells you a story about the audience. A server tells you the audience. The gap between them is where a lot of received wisdom quietly lived.

This is why the internet looks like a more diverse audience than the US TV market ever did. It is not that the audience changed overnight. Direct counting captured the culture panel extrapolation never bothered to sample. Confronted with the truer data, networks are being forced to rearchitect their capital and their investment around who was actually there all along.