Fake test data generator

Test data has a habit of escaping. A notification job pointed at the wrong environment, or a staging system wired to a real SMS provider, will happily message every number in the database. So every number this tool generates comes from a range a national regulator reserves for use in fiction, where no subscriber can hold it.

United States numbers use 555-0100 to 555-0199 in a real area code, the only part of the 555 exchange officially reserved. United Kingdom numbers use the Ofcom drama range 07700 900000 to 07700 900999. Australian numbers use the ACMA fictional blocks beginning 5550. French numbers use the ARCEP mobile range 06 39 98. German numbers use the Bundesnetzagentur drama blocks in Berlin, Hamburg, Munich, Cologne and Frankfurt, which is why German records only come from those five cities.

India has no reserved fictional range. Indian numbers here are format valid and may belong to a real person, and each Indian row is marked as not reserved. Never send a live message to them; route test messages to a sandbox instead.

RFC 2606 reserves example.com, example.org and example.net for documentation and testing. Mail sent to them goes nowhere. Using any other domain for test data, including made-up ones, risks a real person or a real company receiving test emails, and made-up domains are routinely registered later by someone else.

Random test data that cannot be regenerated produces failures that cannot be reproduced. A test fails once with a name containing an apostrophe, nobody saved the data, and the defect is closed as not reproducible. A seed fixes that: the same seed, country and row count always produce identical records, so a failing run can be replayed exactly. Record the seed alongside any failure found with generated data.

The values are chosen to exercise the cases that break real systems. German surnames carry umlauts and the sharp s. French and Australian addresses include apostrophes and accented street names. Several postcodes begin with zero, including Boston, Nice and Darwin, which find every system that stores a postcode as a number. Address order differs by country, with the postcode before the city in Germany and France and after it elsewhere, and the address field contains commas that test CSV quoting.

This is not a substitute for production-like volumes in performance testing. For that, generate data in a script using the same seeded approach, so a load test is as reproducible as a functional one.