Ask a forecaster what went wrong on a bad call and you will usually hear a story about the inside view: a detailed, specific, confident account of this case, built out of this case’s particulars. The story felt like understanding. It was mostly noise.

The fix is older than forecasting and duller than any story. Before you reason about a question, find the reference class and look up how often the answer is yes for questions like it. That number is the base rate, and it is the single most reliable input you have.

Why the boring number wins

A base rate is an anchor you did not invent. Your inside-view estimate is assembled from memory, salience, and whatever you read last night, and every one of those is a bias with a direction. The base rate has none of your recency, none of your narrative, none of your wishful rounding. It is just the frequency, sitting there, indifferent to how interesting your case feels.

The uncomfortable finding, replicated for decades, is that a base rate lightly adjusted beats a detailed model built from scratch, across a wide range of domains. Not because the details never matter. Because you are worse at weighing them than you think, and the base rate already priced most of them in.

The base rate is not the answer. It is the place you are allowed to start arguing from.

Finding the reference class

The whole game is choosing the class. Too broad and the rate is meaningless; too narrow and you are back to a sample of one, which is the inside view wearing a lab coat.

Start wider than feels right. “Startups in this category” before “startups with this exact team and this exact wedge.” You can always tighten.
Prefer classes you can count. If you cannot say how many cases are in the class and how many came out yes, you do not have a base rate, you have a vibe.
Watch for the class that flatters you. If the reference class you reached for happens to make your preferred answer look likely, pick a second one and see if it agrees.

Then, and only then, adjust

The base rate is the start of the argument, not the end. Once it is on the board, you move it: this case has a stronger team, a worse market, a shorter runway. Each adjustment is a push, up or down, and each one should be something you could defend out loud.

The discipline is that the adjustments are visible and bounded. A base rate of 20 percent does not become 80 percent because you feel good about the founders. If your inside view wants to move the number that far, the honest move is to distrust the inside view, not the base rate.

Try it

Open a board, drop a base-rate node on the column you care about, and let it read the frequency straight from the data. Then wire your pushes in, one at a time, and watch the combine hold your estimate to a real probability. The number you land on carries its own argument, right there on the canvas. Start a board.