Concurrent tests without splitting traffic

The usual reason a programme runs one test at a time is a belief that concurrency splits the sample. In Pertento it does not. Each experiment randomises independently, so an eligible visitor enters every experiment they qualify for, and each experiment gets the full eligible audience rather than a slice of it.

Under independent randomisation, each experiment’s own contrast stays unbiased however many others are live, because the other experiments are balanced across its arms by construction. Serialising a roadmap to protect the statistics would be a large throughput loss for no statistical gain.

What concurrency does threaten is narrower and is handled directly. Two variants editing the same element produce a page combination nobody reviewed. A redirect moves traffic into or out of another experiment’s audience and breaks independence. Both are reported by collision detection, and both block automation rather than being left to chance.

Where two experiments genuinely must not overlap, the answer is to stage them in a pipeline. Each stage runs at full traffic in turn, which is the opposite of making them mutually exclusive at runtime and splitting the audience between them.