Why Experiment Calculator works the way it does

Every tool on this site runs in the page. No upload, no account, no server that could be reading your file. That constraint is not a privacy slogan — it is a design decision with costs, and Experiment Calculator pays them.

What it buys: you can drop a CSV with real customer data into it during a meeting and nothing leaves the machine. There is no data-processing agreement to sign, no question about retention, and no answer needed to "where does this go". For anyone who has tried to get a useful tool past a security review, that is the entire value proposition.

What it costs: everything has to fit in a tab, so there are row limits. The CSV parsing is ours rather than a well-tested server library, which means it handles the common cases and will surprise you on an exotic one. And there is no persistence — reload the page and you are starting again, because there is nowhere for state to live.

The trade only works because the datasets are small on purpose. That is not a limitation we worked around; it is the same decision made twice. A dataset small enough to reason about teaches better than one large enough to need infrastructure, and it also happens to fit in a browser.

A tool that requires a cluster teaches you about the cluster.

Experiment Calculator — Sample size before you launch, and a confidence interval after you stop. Both directions, with the assumptions written on the screen rather than buried in a footnote.

All seven tools are on the toolkit page, and every one of them reads the same dataset library.