Blog
Short, practical writing on generating synthetic data that behaves like the real thing — valid checksums, reproducible seeds, and the awkward inputs your code needs to survive.
We used to sell monthly subscriptions with tiers. We replaced them with prepaid credits that never expire. The reason was not the money — it was that tiers forced us to build a product with features deliberately withheld.
Read more →Building the app that runs a German car workshop? You need customers who own cars, cars with plausible plates, and repair orders that belong to the right vehicle. Here is that database, built up one step at a time.
Read more →Correct data, presented in the most disturbing way a real system ever sees it — untrimmed whitespace, invisible characters, homoglyphs, and right-to-left overrides. Does your pipeline survive it?
Read more →Your happy path is well covered. What about the boundary values, the awkward-but-valid records, and the deliberately broken checksums your rejection code is supposed to catch?
Read more →A flaky test that fails once and passes on retry is worse than no test at all. Pass a seed, get the same record every time, and make your fixtures reproducible.
Read more →Random strings that look like a PESEL or a NIP sail straight past your validators — until production data arrives and everything breaks. Here is why format-valid test data matters.
Read more →