Reliable and accurate assessment tools created by experts

Testing Manifesto

The LT123 Testing Manifesto is intended to be a useful document to share information and ideas about assessment. It is a distillation of many years of shared experience within the LT123 team. We hope that it provides a sensible and robust approach to testing, whether thinking in terms of examining, designing tests or delivering them. You are welcome to reproduce it, but please always clearly credit us.

We welcome comments about the content.
You can email us at

The LT123 Testing Manifesto


  • Test developers must define what it is they are testing.
  • Tests must measure what they are intended to measure.
  • Tests must be at the right level.
  • Tests should be an appropriate length.
  • Test tasks must be clear and unambiguous.
  • Test content should be appropriate for the cultures in which it is used.
  • Tests should encourage pedagogically useful preparation.
  • Tests should always be as positive a learning experience for students as possible.
  • Test results should be the same whoever marks them.
  • Test results should be comparable with other tests.
  • Tests must be carefully checked before use.
  • Tests should only be used for the purposes for which they were designed.


  • All testing involves compromise.
  • All testing seeks to generalise.
  • Test questions may not be understood as the test developers intended.
  • Ask yourself what you have learnt about someone if they get your question right?
  • Ask yourself what you have learnt about someone if they get your question wrong?
  • Tests reveal what someone knows and doesn’t know now, not what they have known or will know.
  • Consider the balance between testing knowledge and testing what learners can do.
  • A good test can be poorly delivered.
  • A good test can be poorly deployed.
  • A good test construct can be turned into poorly designed tasks.
  • Very precisely focused questions may neglect the bigger picture.
  • Repetitional exercises may be desirable in learning but they become interdependent (and thus unreliable) in testing.