Guides
Self-Hosting the Output: A Practical Checklist Before You Trust Code Export
A code export button is not the same thing as a project that runs somewhere else. Brief six in our testing exists because we kept finding the gap between the two. Here is what to check yourself.
Tom Brackett ·
Code ownership is the axis we weight lowest, at three percent, and it is also the axis we get asked about most in correction requests, usually from a reader who exported a project, tried to run it somewhere else, and hit a wall the marketing page did not warn them about. The low weight reflects that most projects never leave the platform they were built on. The frequency of the question reflects that when it matters, it matters a great deal.
What brief six actually tests
Our sixth fixed brief takes the finished application from brief one, exports it using whatever mechanism the tool provides, and tries to run it in two places: on our own development machine, and on a plain container host with none of the platform's own infrastructure available. We record whether it starts at all, and if not, exactly what it was missing.
The failures cluster into four categories, in order of how often we see them.
Hidden platform services. The exported code imports a client for a database, a queue, or a file store that only exists inside the original platform. The code is real and readable, but it will not run anywhere else without a rewrite of the data layer.
Undocumented environment variables. The project needs a specific set of secrets and configuration values to boot, and the export does not include a list of what they are or what they should contain. You are left reverse-engineering the requirements from stack traces.
Build tooling that assumes the platform's own pipeline. The export is source code, not a buildable project. It references a build step, a bundler configuration, or a deployment script that lives on the platform's servers and was never included in what you downloaded.
A codebase nobody could maintain. The project starts, technically, but the generated code is structured in a way that assumes an AI agent will always be the one editing it. Variable names, file organisation and comments are optimised for a model's context window rather than a human's, which matters the moment you need a developer who is not using that platform to change something.
What to check before you rely on code ownership
Do not wait until you need to leave to find out whether you can. Export a small test project early, before you have invested months in the platform, and try brief six yourself: run it somewhere else, with no help from the original vendor's infrastructure. If it does not start, you now know the real value of "own your code" on that tool's pricing page, and you know it while switching still costs you an afternoon rather than a migration project.