FastAPI: 7 of 8 endpoints broken, all defects already diagnosed

FastAPI backend, eight computational engines, one shared code path building every HTTP response. It works for one engine and fails for the other seven. Of those seven: three return HTTP 200 with a correctly structured response in which every data field is null. Four return HTTP 500. Each engine declares its own output contract — the name, type and layer of every field it emits — and the requirements for its input data. The shared code path does not read those declarations. That is why it succeeds for one engine and fails for seven. You receive: repository access, deployment access, a document listing fourteen defects with the file, line and fix identified for each, and a reference export for every engine. Completion: one script covering all eight endpoints. For each endpoint it asserts that the data layer is not entirely null, that the response field list matches the engine's declaration exactly, and that the output matches that engine's reference export. The script exits 0 on my infrastructure. Out of scope: the mathematics, the engines themselves, the registry, new abstraction layers, renaming. Stack: Python, FastAPI, pandas, parquet, Docker, nginx, Linux.