A cognitive bias is not a slip. A slip is a thing you would take back if it were pointed out. A bias is a shape the judgement takes anyway, in a direction that can be written down before you have made it, by someone who has never met you. It is not a shortage of effort or care, and more of either does not remove it.
The one below is anchoring. Show somebody a number before asking them to estimate something, and the estimate moves toward the number. Tversky and Kahneman had people watch a wheel of fortune stop on 10 or on 65, then asked what percentage of the countries in the United Nations were African. The group that saw 10 said 25. The group that saw 65 said 45. The wheel was rigged and everybody watched it spin.
So: six questions. Before each one you are shown a number and asked whether the answer is more or less than that. Three of those numbers sit far above the true answer and three sit far below it, and which question got which was decided by a coin flip when you pressed start. Then you give your estimate.
Here is the problem with running this on you. Anchoring is a claim about two groups of people, not about one person. The finding is that a group given a high number answers higher than a group given a low number. You are one person. You cannot be both groups, you have no untouched version of yourself to be compared against, and any estimate you make is one number with nothing to hold it up to.
What six items can and cannot do
There is one thing the run can do honestly. The anchors were assigned at random, so the three high-anchored questions and the three low-anchored ones form a real randomised comparison — just a tiny one, and across different questions rather than different people. The page ranks your six estimates by how far each sits from its true answer and asks how likely that arrangement would be if the coin had meant nothing.
The arithmetic caps out immediately. There are twenty ways to split six items into three and three. If all three of your high-anchored estimates land above all three low-anchored ones — the strongest possible outcome — that is one arrangement in twenty, p = 0.05, and that is the floor. Six items cannot produce a smaller number no matter what you do. Anything short of a perfect separation cannot reach it at all.
So the result is a shape and not a finding, and the page says which one it got. What it will not do is take one person's six answers and report an effect that took Tversky and Kahneman groups of people to establish.
What it costs you
The first number said out loud sets the range. Whoever names a figure first in an estimate, a budget or a deadline has moved everyone else's answer, including the answers of people who think they are arguing against it.
Every default is an anchor. A pre-filled amount, a pre-selected plan, a suggested tip. The ones people choose instead are still measured against the one you put there first.
Knowing about a bias does not switch it off. The wheel was visibly rigged and it worked anyway. Anchoring is not credulity, and a warning at the top of the page is not a fix.
Removing an anchor is also a decision. An empty field with no suggestion does not leave the judgement clean. It hands it to whatever number the person happened to see last, which is one you did not choose and cannot see.
I picked six questions with answers I had to check, and while writing the anchor values I kept drifting toward numbers near the true ones because they looked more plausible on screen. They are wider now, and I still do not know whether the ones I first wrote were a design instinct or the same effect this page is about.