Tversky and Kahneman (1974) demonstrated that arbitrary numbers, even ones people know to be random, powerfully distort subsequent estimates. In one version, participants spun a wheel rigged to land on 10 or 65, then guessed the percentage of African countries in the UN. Those who saw 65 guessed 45%; those who saw 10 guessed 25%. This simulation runs an analogous scenario on 200 virtual agents estimating the price of a "Golden Chronometer" (true value: ₹50,000) after the wheel produces a random anchor.
Each agent has a Prior Knowledge score (0–100%) drawn from a normal distribution. When an anchor is introduced, each agent's estimate is computed as a weighted blend of the true value and the anchor, with the weight determined by their knowledge — plus calibrated noise. Low-knowledge agents weight the anchor heavily; high-knowledge agents resist more but are never fully immune. A D3.js histogram shows the full distribution shifting in real time, with the pre-anchor "ghost" preserved for comparison.
The scatter plot is the clearest view of the mechanism: plot Prior Knowledge on the X-axis and Estimate Error on the Y-axis. You see a gradient — low-knowledge agents (left) cluster far from zero, while high-knowledge agents (right) cluster near it. But even the most knowledgeable agents rarely land exactly at ₹50,000. The adjustment gap persists even when people know the anchor was random. That's the core horror of anchoring: awareness doesn't cure it.