Principle
Prefer a test that kills hypotheses
A test that can only confirm your favorite story is a weak detective.
When it fits
- The investigation has accumulated several plausible stories and more logs are not narrowing them.
When to avoid it
- Operational tests may be confounded or only suggestive; preserve uncertainty when the result is not decisive.
Why it matters
Choose the next test for discrimination: ask which result would make one or more hypotheses much less plausible. Write the competing explanations before the test and predict what each would produce. A useful negative result can be progress because it shrinks the search space.
An example
Use the application's own database credentials to distinguish an authorization failure from a more general connectivity failure.
Check your result
Before running the test, you can state at least one result that would change what you investigate next.
Keep this limit in mind
- Operational tests may be confounded or only suggestive; preserve uncertainty when the result is not decisive.
Connected ideas
Useful withOrder troubleshooting tests by likelihood and risk
Evidence and sources
Google SRE advises designing tests that can rule groups of hypotheses in or out when possible.
Many operational tests are only suggestive; uncertainty should remain visible rather than being forced into a binary verdict.
Effective Troubleshooting · Test and Treat; test design considerations