A prototype is a question expressed in code. Its value comes from the answer, not from surviving forever.
Write down the question
Before building, I try to name the uncertainty: can this interaction feel good, can this service meet the latency target, or can someone understand the workflow without instruction?
Preserve learning, not accidents
If the experiment succeeds, keep the decisions, measurements, and constraints. Rebuilding the implementation is often cheaper than carrying every shortcut into production.