Generate Consistent Outputs for Similar Inputs

Design a system that generates an output for an input and returns the same output for inputs considered similar. The input domain, output-generation process, and definition of similarity are unspecified. Start by explaining the contract you need to establish, then propose a design under clearly stated assumptions. Do not assume that similarity means an embedding distance or that output generation requires an ML model. ### What a Strong Answer Covers - A precise distinction between identical input, similar input, and equal output. - Whether the similarity rule supports stable groups and how conflicting matches are resolved. - A path for finding an existing output or creating and publishing a new one. - Consistent behavior when similar requests arrive concurrently. - Versioning and recovery when the similarity policy or generator changes. ### Follow-up Questions - If A is similar to B and B is similar to C, must all three receive the same output even when A is not directly similar to C? - How would a policy update that merges two existing groups affect their previously returned outputs?