StackPilot Pro · Notes & guides

Before you call an AI prototype finished

A polished prototype answers a useful question: can someone understand and operate this flow? It may leave other questions open. Does the data persist? Does the real integration respond? Do permissions hold for another account?

Use a short evidence record so the next person can see where the work stands.

Walk through the task

Choose the main action a visitor came to perform. Follow it from its entry point to its result. Read every label as if you had never seen the project. Check the links, empty states, and the route back when something goes wrong.

For a checklist generator, that means entering a brief, producing the checklist, changing an answer, and downloading the updated result. A button that changes its appearance has not yet proved the download works.

Name the environment

Write down where each check ran. A local mock, a staging service, and the live provider establish different things. If an integration was simulated, say so next to the result.

Keep evidence proportional to the change. A small copy edit may need a readback and a narrow visual check. Shared storage, permissions, or payments need checks along the actual dependencies.

Try a useful failure

Choose a failure the flow can realistically encounter: a missing field, an unavailable dependency, a duplicate submission, or an empty result. Confirm that the visitor can understand the problem and recover without losing useful work.

Do not introduce real charges, production writes, destructive actions, or extra access simply to test a page. Establish the permitted environment and actions first.

Check the screen people will use

Try a narrow screen and keyboard navigation. Check that text stays readable, controls remain reachable, focus is visible, and long input does not push actions off the page. Review the open state of menus and selectors as well as their closed appearance.

Write the limits beside the result

“The preview renders” is useful evidence. “Ready for production” is a broader claim. State the specific checks that passed, what remains untested, and what that means for the next step.

AreaEvidence to record
Main taskThe steps performed and the observed result
DataWhere it is stored and whether persistence was checked
IntegrationReal or simulated; environment and response observed
AccessAccounts and permission cases actually exercised
InterfaceScreen sizes, keyboard flow, and failure state checked
Remaining workThe missing dependency or decision and its implication

The goal is a handoff someone else can assess without reconstructing the conversation.