Field note · February 28, 2026
A five-minute use for Distribution Lab
Sample from seven distributions, watch the central limit theorem happen at whatever sample size you choose, and see exactly when the mean stops being a useful summary.
The fastest way to get value out of Distribution Lab: pick a dataset from the library, open the tool, and answer one question you would otherwise have written a notebook for.
That is the entire pitch. Not a replacement for your actual environment — a way to check something in the ninety seconds before a meeting without setting up a kernel and remembering which virtualenv had pandas in it.
The questions it is good for are the ones with a yes-or-no shape. Is this column skewed. Do these two things move together. Does the split look even. Is the tail long enough to matter. Each of those is thirty seconds here and ten minutes anywhere else, and the ten-minute version is why they routinely go unasked.
The questions it is bad for are anything you need to repeat. There is no saved state, no scheduling, no way to hand someone a link that reproduces what you did. That is deliberate — a tool that quietly becomes production infrastructure is a worse outcome than one that cannot.
The bottleneck in most quick analysis is setup, not compute. Removing it changes which questions get asked at all.
Distribution Lab — Sample from seven distributions, watch the central limit theorem happen at whatever sample size you choose, and see exactly when the mean stops being a useful summary.
All seven tools are on the toolkit page, and every one of them reads the same dataset library.