A classification prompt carries four worked examples. Three are laid out as "Input:" and "Output:" on separate lines; the fourth uses a dash and puts both on one line. Results are inconsistent in shape. What does this indicate?

The principle — Examples teach by pattern, and layout is part of the pattern whether you intended it or not. A set that disagrees with itself about form teaches that form is negotiable.

Why the key is correct — Google states that one of the primary objectives of adding few-shot examples is to show the model the response format, and that it is therefore essential to ensure a consistent format across all examples, paying attention to tags, white space, newlines and example splitters. Three examples in one shape and a fourth in another supply two competing templates, so the output shape becomes a coin toss. The fix is to reformat the fourth, not to delete it.

Why the others are wrong — There is no fixed ceiling on example count, and deleting one throws away its coverage. Overfitting produces answers that hug the examples, not inconsistent layout. And nothing here indicates a stated formatting instruction for the fourth example to contradict.

Remember this — Vary what your examples say. Never vary how they are laid out.

Sources — Google, Prompt design strategies.