The pipeline runs in the customer’s own DEV environment, on the customer’s own infrastructure, against the customer’s own data.
The layers
- Adversarial specification review — the specification is attacked before any code is written against it.
- Golden-record behavioural replay with mandatory per-divergence classification. Every difference in behaviour must be classified. None can be waved through in bulk.
- Executable acceptance criteria, so “done” is a thing that runs.
- Characterisation testing for the behaviour nobody wrote down but everybody depends on.
- Your own scanners, in the pipeline rather than beside it.
The honest part
The golden corpus and the divergence-classification questions are the hardest problems in the product and remain unproven. No amount of architectural cleverness affects them, and they are the basis on which this should be funded or declined. We would rather say so here than have you discover it in month four.