Vibe coding lets anyone build a web app with a natural language prompt, promising you'll never read generated code. Yet silent failures remain a major issue. The UI may look fine, but clicking an 'Add to Cart' button might show a success message and update the cart count without actually saving data to the database. Debugging requires interacting with the UI, reading logs, and ultimately reading the generated code, defeating the purpose of vibe coding.

A study of vibe-coded real-world apps found that even frontier models frequently introduce silent failures across iterations. These include failures in tracking state updates, cross-handler state disconnects, and disconnected UI feedback—where a success message appears but data isn't saved. For example, adding a promotional-code feature might display 'Discount applied!' without persisting or showing the new total.

Current verification methods are inadequate. LLMs as judges are unreliable, often hallucinating or missing bugs. Unit tests require writing more code, lack UI-to-backend integration coverage, and are inaccessible to non-programmers. Static analysis is accurate but has a steep learning curve.

Flow Check, developed by a Ph.D. student at Columbia's DAP Lab, is a constraint language and static analysis pipeline that lets users specify how an app should behave directly from the interface. Users select UI components, APIs, and storage via an overlay, expressing constraints like 'When I take [action], these update: [component],' then checks them against the actual code without reading a line.