Acceptance criteria are often written as a checklist to tick off rather than a claim someone might dispute. The second kind is more useful and noticeably harder to write.
The tell
If every criterion on a ticket can be verified by reading the code once, the criteria were probably written after the implementation, in its image. Good acceptance criteria should be writable — and arguable — before anyone has seen the solution.
What we ask for instead
- A claim, not a task — "the user cannot submit a form with an expired session" rather than "add session check to form submission".
- A specific counter-example the tester should try — the criterion should suggest its own attack, not just its own happy path.
- A stated boundary — what happens exactly at the limit, not just comfortably inside or outside it.
Why this changes what testers do
A checklist criterion invites confirmation: does the button appear, yes or no. A claim invites disagreement: is this actually true, and what would prove it false. Testers given claims find more, because they are being asked to look for the counter-example rather than confirm the happy path.
A criterion nobody could disagree with was probably not worth writing down.
Where this collides with generated tickets
A model given a rough feature description will often generate acceptance criteria that read fluently and satisfy the letter of "has acceptance criteria" without containing a genuine claim. We now review generated criteria specifically for this — restating the implementation instead of stating a requirement — and rewrite anything that fails the test of being disputable.
The habit that fixed most of it
Whoever writes a ticket's acceptance criteria is not allowed to also be the first person to test it. A different pair of eyes, unfamiliar with how the criteria were intended, reliably finds the gaps the author could not see, because the author already knows what they meant.
What it bought us
Fewer tickets marked done and reopened a week later for behaviour that technically matched the criteria but obviously missed the point. That gap — technically correct, obviously wrong — is exactly what disputable criteria are meant to close.