The Score That Never Resolves
A president's approval rating moves every week for four years, and nobody ever finds out whether last Tuesday's number was right.
The number itself has a built-in problem, regardless of how carefully anyone builds it. The approval average, the rolling line built out of a dozen surveys with different house effects, smoothed and weighted into one wiggling curve, has nowhere to go on the calendar to get checked. It isn't a forecast of anything, just a running tally of how many people, when asked this week, said yes instead of no. Next week it updates, older responses age out, and the line moves again. There is no day when a fact from outside the poll shows up and says the line was three points off.
Compare it to the number sitting right next to it in the same news cycle: the win-probability line a forecasting model puts on a candidate ahead of an election. That number is built the same way, aggregated polls, adjusted, weighted, smoothed into a curve that wiggles for months. But that curve has an appointment. On election night, a number arrives from outside the polling apparatus entirely, actual votes counted at actual precincts, and the forecast gets marked against it. Wrong models get named. Right ones get cited for four more years. The mechanism is identical to the approval tracker's mechanism. The only difference is a date on the calendar when the curve stops being an internal aggregate and starts being a graded prediction.
That date is the entire difference between measuring something and recording an argument. Take it away and the two curves are indistinguishable from the outside. Both wiggle in response to news. Both smooth out short-term noise into an apparent trend. Both can narrow or drift in ways that look exactly like convergence on a fact. But the approval curve isn't converging on a fact, because there's no fact waiting for it. It's converging, if it converges at all, on consensus: respondents anchoring their answers to what they already believe most people think, and models tuned to agree with each other because disagreement gets punished before any election proves anyone right. The stability isn't evidence the number found the truth. It's evidence the argument settled into a majority and a minority, and the score is just reporting the current head count.
None of this requires bad faith from anyone building the tracker. A rolling average with no resolution date will still look, week to week, exactly like a rolling average with one. The smoothing math doesn't know the difference. Only the outside world does, and only on the day it shows up. The honest question to ask about any moving score isn't whether it's well constructed. It's whether there's a day circled on a calendar somewhere when something outside the scoring system gets to overrule it. If there isn't, the wiggle is real and the trend is real, and none of it measures the thing it claims to measure. It measures which side of an unsettled argument currently has more votes, updated continuously, forever, by design.
Approval ratings are the clean case because the resolution date used to exist and got removed. A president is elected once, sometimes twice, and in between, the number that claims to track how the public feels floats free of any check for years at a stretch. It rises through a good quarter, falls through a bad one, and every move gets reported as if it landed on something. It hasn't. It's waiting for a Tuesday in November that may not even test the thing it was supposedly measuring, since a national vote and a national mood are not the same fact. When that Tuesday finally comes, it resolves the election. The approval line, still wiggling, goes right on not resolving anything at all.