Sampling reduces execution volume; it does not prove that untested content is accessible.
Population inventory
Before selecting a sample, record the full known population of pages, views, documents, templates, component types and processes. Include:
- common pages, regions and components;
- essential functionality and critical transactions;
- unique templates and content types;
- languages and locale variations;
- desktop, tablet and mobile responsive variations;
- public, authenticated and role-dependent states;
- loading, empty, validation, interruption, recovery and confirmation states;
- every complete process required to accomplish an essential task;
- known conforming alternate versions and their discoverability.
Unknown population size is a limitation, not permission to describe a sample as complete.
Structured sample
Select representative targets for every inventory dimension. Record the selection rule, target identifiers and why each target represents a population group. Include high-risk and historically defective content even when it is uncommon.
Random sample
Where random selection is used, preserve the population list, method, seed or randomisation record, selection date and selected targets. Random sampling supplements structured sampling; it does not replace essential processes, common components or unique content.
Complete processes
Test complete critical processes from entry through success, including validation errors, interruption and recovery. A passed isolated component does not establish that the process works.
Freeze and change control
Record the product build, content snapshot, sample version and applicable environments before execution. Added or removed targets require a change record stating the reason, effect on coverage and whether earlier results remain comparable. Do not silently replace a failed target.
Coverage calculation
Report both the known population and executed count for targets, component types, states, responsive variations, environments and complete processes. State exclusions and unresolved inventory gaps. A percentage is meaningful only when its denominator and selection method are recorded.
Result language
Say “no failure was found in the executed sample” when that is what the evidence supports. Do not say “the product passes” or use “every” unless the complete known population and all required states were actually evaluated.