“The dashboard works” is not an acceptance criterion. A founder needs a repeatable way to verify that a user can complete the promised task and that the stored result matches what the interface claims.
Write acceptance criteria as examples with a setup, action, expected outcome, and evidence. This makes review possible without requiring the founder to judge implementation details they cannot inspect confidently.
Define the result at the workflow boundary
Choose behaviours that matter to the user: a response is received, an approved record is exported, or a payment grants the intended access. Include the backend result when the interface alone could be misleading.
Avoid criteria that prescribe a tool unless that tool is an actual requirement. “Uses a background queue” describes an implementation choice. “A slow export remains visible and can be downloaded after completion” describes an outcome the founder can test.
Write a fictional study-collection acceptance set
For a customer-study MVP, the first example might read: given a published study with two questions, when an invited participant submits text answers, the founder can see both answers under that study after signing out and back in.
Add a privacy example: given two founder accounts, when the second account opens the first account's response URL, it cannot read the private response. Add a lifecycle example: when the owner closes a study, a new participant receives a clear closed-study message and no new answer is saved.
Add a recovery example: if submission fails before storage, the participant sees a retryable state; if storage succeeded but confirmation was interrupted, retrying must follow the agreed duplicate rule. The expected behaviour needs an explicit product decision before acceptance, not an improvised judgement during the demo.
These examples are proposed criteria. They do not claim that a particular product has passed them or define a universal level of reliability.
Agree who supplies the evidence
Ask the developer to provide a review account, fictional test records, a short test script, and any relevant exported or stored evidence. The founder should be able to repeat the core checks in the review environment.
Separate defects from changes. If a response disappears despite an agreed persistence criterion, that is a failure against the scope. If the founder now wants multilingual branching questions, that may be a new requirement. Clear examples make that distinction easier and reduce arguments about what “complete” meant.
Acceptance checks have limits. Passing a few cases does not establish load capacity, accessibility, or every permission boundary. List additional reviews appropriate to the pilot rather than claiming the examples prove production readiness.
The StackFlight case study provides a delivered study-workflow reference, while your own criteria should reflect your actual product and operating promises.
Questions about founder acceptance
Should the founder test every technical component?
No. Verify the business outcomes and request appropriate engineering evidence for deeper checks. Keep ownership of both kinds of review clear.
Can a recorded demo count as acceptance?
It can support review, but repeatable access and test data provide stronger evidence than a recording alone. Agree the method in advance.
Where should the criteria live?
Keep them with the scope and link each milestone to the relevant cases. Bring one ambiguous criterion to a product scope review before development begins.