Blog

Writing about shipping software faster than you can check it: what quietly breaks, how to catch it yourself, and what we have measured about our own tool when the number is worth printing.

Product

A confirmed bug becomes a standing check

Every tool can generate tests. The hard part is knowing which of them could ever fail. A re-check is not counted as protection until it has been seen to do it.

5 min read

How the persona panel picks its testers

Six simulated testers, drawn from ten thousand records, matched to the audience you describe. How the draw works, what it refuses to model, and what it is not.

6 min read

All product posts

Research

All research notes

Articles

36 out of 36 is not 100%

The formula in every textbook says a rate can be negative, that 19 of 20 is 105%, and that a perfect score carries no doubt at all. At 16 trials its real coverage drops to 3.9%. The fix is one line.

7 min

Which accessibility issues can be automated?

An automated checker ships around a hundred rules. Counted against the spec they land on 29 success criteria, and 29 of those rules cover one criterion between them.

7 min

You are not your user, in five numbers

Ten thousand simulated users from a research dataset. 27% write code, 12% are digital natives, and only one in five arrives trusting your software.

6 min

How to tell a website was built by AI

There is a house style now, and once you have seen it you cannot stop seeing it. Seven tells, in the order you meet them. Every one of them is something a person can see, read or press, with no developer tools at all.

8 min

All articles