A thin cell is a group you barely reached: say the handful of young renters in one region who happened to fall into your sample. You can splice in extra answers generated by an AI model to fatten it out. The first slider sets how many real people you actually reached. The second sets how many synthetic answers you pour in on top. Watch three things move: how tight the estimate looks, and how far off it really is.
Drag left for a genuinely thin cell, right for one that was almost fine already.
The boost knob. The same knob fills a thin cell honestly or manufactures a number.
The honest starting point: noisy, but nothing invented.
The number a buyer sees. More rows, tighter bar, every time.
Once the model's flattening is counted. The gap below the green bar is the lie.
Move a slider to begin.
Illustrative simulator, not a measured curve. The draft is explicit that no published dose-response curve exists yet, so this reproduces the documented shape only: big honest gains on the smallest cells, gone once a cell reaches roughly 150 to 200 real people (Dig Insights, 2026), and an interior best mix beyond which added synthetic answers narrow the estimate around the model's guess (Huang, Wu and Wang, 2025). A Webflow embed cannot call a model. Shape recorded 2026-08-18.