Four people in Bristol.

We spent long enough waiting on other people's test suites to be reasonably sure the waiting was avoidable.

Plant room ceiling, ranks of fans and their motors receding under a low roof.

The name

A bellwether is the sheep the shepherd puts the bell on. It is not the fastest or the strongest one. It is the one whose position tells you where the rest of the flock is, which means you can stop counting sheep and start listening.

That is the whole product argument. A suite of four thousand tests contains a much smaller set whose result tells you what the four thousand would have said. Finding it is a mechanical problem, not a judgement call, and it is a different set on every commit.

What we will not do

We will not predict which tests are likely to fail. A prediction that is usually right is worse than no prediction at all, because a suite you cannot trust is a suite you stop reading. Bellwether skips a test only when it can name the recorded fact that makes the skip safe.

We will not sell a seat tier that withholds selection, we will not bill for machine minutes, and we will not put a number on your dashboard that we cannot show the working for.

Who

Bellwether is built by Ruth Okonkwo-Vance, Tomas Bergqvist, Ada Fennimore and Serge Iliev. Between them: eleven years on build infrastructure at a bank that will not be named, two compilers, and one very long argument about whether coverage is a measurement or an opinion.

We answer the phone. 020 7946 0347, or [email protected].

Read the skip list on your own repository. That is the only demo worth having.

Start a trial