Certance Research · The Standard · Edition 2026.6
Quality is the evidence you can hand to someone else. The Certance Standard defines how software teams keep that true when AI agents write most of the code and AI runs inside the product: five merge gates, numbered rules and twelve statements.
Twelve statements: the five principles of Chapter 1 (1 to 5) and seven requirements the later chapters develop (6 to 12).
1. AI is an amplifier of your engineering system. Harden the system before scaling the agents.
2. Every merge rests on behavioural evidence. A green pipeline shows only that the tests passed, and the tests may have moved with the change.
3. Every AI-assisted change carries its own proof and a named owner. Five hard gates on each: provenance, test integrity, behavioural evidence, contract check, blast-radius match.
4. Every change declares its provenance. Every change traceable to its author, human or agent, with the access it held at the time.
5. People keep the judgement. Humans scope the journeys and design the coverage; agents implement. "Seems fine" is a stop signal.
6. Production is cheap; verification is the asset. Size your process for the review queue.
7. Most of the cost of code comes after it is written. Accept generated code only at a maintenance cost the business can carry.
8. The specification is an artefact. The human owns what must be true; the agent owns the implementation; a model never validates its own work.
9. Agents are governed infrastructure. Isolation, least privilege, configuration as code, output reviewed like third-party code.
10. Quality is the evidence you can hand to someone else. A colleague, a customer or a supervisor must be able to check it.
11. Evidence must outlive the sprint. Stored, linked to its release build, and kept for a retention period the firm sets and records.
12. AI inside the product is tested with the journey it serves. Evaluation sets are protected tests, pass rules hold over repeated runs, and every model change goes through the gates.
For what these principles change in a testing team, see quality assurance vs quality engineering.
The Standard is the point of view. The practice that applies it to client teams is Certance Quality Engineering.
The Standard, edition 2026.6, sets out numbered rules across fourteen chapters: five hard gates for every change, the testing levels for code agents write, continuous integration and release gates, running an agent roster, testing AI inside the product, evidence for supervisors, and when a release is tested enough. It is 42 pages and free to download. Its companion, Evidence and Guidance (34 pages), explains why each rule exists: the evidence behind every chapter, the EU and UK regulatory overlay and a plan for adopting the Standard. Leave your details below to download it. If your team wants the Standard applied to its own suite, that is the Coverage & Evidence Audit.
Download the Standard (PDF, 42 pages)Get Evidence and GuidanceExecutive Briefing (4 pages)The companion to the Certance Standard, edition 2026.6: 34 pages, PDF. Leave your details and the download opens straight away.
Edition 2026.6 was checked against its primary sources in October 2026. The Standard lists the sources for its rules; the full list, with the evidence behind each chapter, is in Evidence and Guidance, under Sources and references.