Retrieval changes what a model reads. Fine-tuning changes the model itself: its weights move so that it produces the shape of output you showed it. That difference is the whole decision between the two, and it is why a fine-tuned model can be wrong on facts it never saw.
About 1 min of reading
State the job in one line.
The job
Before you run it, say out loud what it will print. Then open the answer.
Read the first example one line at a time. A line that prints shows what appears on screen; the rest set things up quietly.
Write the one thing the tuned model must do: the format, the tone, the classification, the transformation. If you cannot name it in a sentence, you do not have a tuning job yet.
A tuned model is a version, not a replacement. Keep the base available so you can compare, and so a bad run can be reverted rather than argued about.
Show a model two hundred examples of your format and it learns the format well. Show it facts and it learns them loosely, and it will rephrase or drop them. Style and structure are what tuning is good at.
A tuned model is a build artifact. It does not update when your documents change, which is why a model tuned on last year's prices answers from last year's weights.
Tuning takes a training run over your examples, and the result is a model you then serve. That is a different cost shape from retrieval, which pays per question.
Each one is defined here on this site. Open the entry for the full definition and the other words it points at.