The ride was fine. The driver took a puzzling route, the car smelled of synthetic pine, and nothing actually went wrong — so when the app asks how it went, your thumb does what it always does. It hovers over four stars, good-not-great, and then taps five. Not because the trip was excellent, but because you know what four does. On a platform where a driver’s income hangs on his average, four is not a grade. It’s a penalty. You have essentially been asked whether a stranger should keep working, and the ride wasn’t that bad.
Multiply that small act of mercy by every passenger, diner and buyer on the internet and you get the modern rating scale: not a measurement but a switch. Five means “do not harm this person.” One means “something genuinely terrible happened.” The three options in between are a dead zone, avoided precisely because they still look like honest feedback — and honesty, once a number has consequences, reads as cruelty.
The strongest evidence that the inflation is about consequences rather than quality came via NPR in 2019, in a study of a platform with more than a billion dollars in transactions. The researchers — Horton, Golden and Filippas — found that when users were told their feedback would stay private and would not be used to punish the provider, they gave much more critical feedback. Same work, same scale; the only variable was who would get hurt. Horton’s summary of the typical platform: it grades “with an A as the average grade.”
The ceiling is visible once you know to look for it. An analysis by Rankquant, a review-analytics vendor that stresses its numbers are directional, found the typical listing parked above the midpoint on every major surface it checked — Amazon around 4.4 stars, Yelp restaurants about 4.2, literary fiction on Goodreads above 4.1, Booking.com hotels near 8.4 out of 10. And fear for someone’s livelihood isn’t the only gravity at work. When the data-visualization company Datawrapper compared more than two billion ratings of some 4,000 films across IMDb and Letterboxd — a context where no one gets fired — the midpoint of each scale drew just 6 percent of votes, while the option at 80 percent of the scale (8 out of 10, four stars) was the most common, taking roughly a quarter. Some of that is self-selection — people rarely finish films they expected to hate — but that only widens the point. Any social pressure bends a rating upward. Employment pressure just bends it hardest.
Once everything piles into the 4s, the ranking still has to happen, so the real discrimination moves into territory no rater ever chose. Nobody has ever tapped 4.63. The interface gives you five buttons; the marketplace runs on hundredths. The designer Tobias van Schneider has described the five-star scale as “four different ways to say ‘this sucks,’” a thin band from 4.1 to 4.9 that says “this is good,” and “a way to say you are a liar” at a perfect 5.0. It was a joke. It reads like an engineering spec.
A scale this collapsed survives because it still looks like data. It produces averages and rankings; it lets a platform claim its marketplace is sorted by customers like you; it converts the one thing software cannot process — your mercy — into a number filed under someone else’s name. Five stars no longer means excellent. It means “that was fine, and I don’t want anyone fired over it,” which is a fair description of most transactions on earth and has never once been offered as a button. Everyone agreed. No one was asked.