There is a belief in many teams that acceptance criteria are a formality: a few bullet points under a user story that the product owner types so the ticket looks complete. Write something like “login works as expected”, tick the box, move on.

That belief is wrong, and expensive. Acceptance criteria are the only place where the team agrees, before building, what “done” means for this story. If a criterion cannot fail, it cannot tell you anything. Let us take one apart.

The login story, as usually written

The story: “As a registered customer, I want to log in so that I can see my orders.” Typical criteria underneath:

  • User can log in successfully.
  • Invalid credentials show an error.
  • The page should be secure and fast.

Try to imagine a build that fails each one. “Successfully” compared with what? “An error” which error, shown where? “Secure and fast” has no failing state at all; any build can be argued to satisfy it. Three developers would build three different things, and all three would claim to meet the criteria. That is the test of a bad criterion: does it settle arguments, or start them?

Myth: more detail means more bureaucracy

The second belief is that making criteria specific slows things down. In practice the opposite happens, because the conversation you skip now happens later, in a bug report, with more people involved.

Rewrite the login criteria so each has an observable condition and an outcome. Many teams use the Given-When-Then shape because it forces that:

  1. Given a registered, active customer, when they submit the correct email and password, then they land on the Orders page and see their name in the header.
  2. Given a registered customer, when they submit a wrong password, then the page stays on login and shows “Email or password is incorrect” without saying which one was wrong.
  3. Given five consecutive wrong passwords for one account, when a sixth attempt is made within 15 minutes, then the account is temporarily locked and the message says to try again later.
  4. Given a customer whose account is deactivated, when they submit correct credentials, then they are refused with the message “Account inactive. Contact support.”
  5. Given a login page served over a network, the response time at the 95th percentile is under 2 seconds for up to 200 concurrent users.

Every line can fail. Each also exposed a decision someone had to make: the lock threshold, the wording that avoids confirming whether an email exists, what happens to inactive accounts. Those decisions were going to be made anyway. Either the product owner makes them in refinement, or a developer makes them silently at 11 pm. Neither is wrong in principle; the silent one is just invisible until it is a defect.

Myth: criteria are the test cases

Third belief: if the criteria pass, the story is tested. No. Criteria describe the agreed behaviour at a coarse grain. They are examples chosen to make the requirement concrete, not an exhaustive list.

Criterion 1 above is one example of a valid login. A tester turns it into several: the password with special characters, the email in different letter case, leading and trailing spaces, the maximum password length, a very slow connection, the back button after logout. Criterion 3 hides boundary questions: is the lock after the fifth failure or the sixth? Does a successful login reset the counter? Does the lock expire on its own? The criteria give the team a shared floor. Testing goes above it.

Likewise, criteria written by the product owner alone will miss things only a tester or developer notices. This is where a tester earns their place in refinement. Reading a story and asking “what happens if…?” is cheap, and every answer either becomes a new criterion or is explicitly ruled out of scope.

Rules of thumb for checking a criterion

No checklist is universal, but these questions catch most weak criteria:

  • Could two people watch the same demo and disagree on whether it passed? If yes, it is vague.
  • Does it contain words like “fast”, “user-friendly”, “appropriate”, “etc.”? Replace them with a number, an example or a reference.
  • Does it describe behaviour the user or system shows, not how the code is built?
  • Is it about one thing? A criterion with “and” in the middle often hides two.
  • Is there at least one negative or edge case, not only the happy path?

Non-functional criteria, such as the response time in item 5, need extra care. They need a measurement condition, here a user count and a percentile, or nobody can say whether they were met.

One last point against over-correcting. Ten pages of criteria for a small story is its own failure, because nobody reads them and they go stale. Aim for the handful that capture the decisions people might otherwise make differently. Where a rule is already stated elsewhere, such as a password policy document, link to it rather than copying it.

Leave a Reply

Trending