01
The problem
An external service can fail or return incomplete data. The interface should explain the failure and make recovery possible without inventing results.
02
The decision
Send bounded text in a request body, validate all five scores and preserve the input after a failure. An older response cannot overwrite a newer edit.
03
The evidence
API and browser tests cover results, network failures, invalid responses and edits during a request. Automated browser checks use mocked responses, separate from real inference.
04
The context
An IBM/Coursera course project. Watson provides the model. Scores do not diagnose emotions or clinical conditions; the app asks for fictional English examples.
