Few-shot Prompts for SL-CAI (Appendix E.1) — prevents role confusion in → Critique and Revision

critique-and-revision asks the same underlying model to play two different roles in sequence within one context — first critic, then reviser — and this repeatedly caused a specific bug: "the language model sometimes becomes confused about its point of view — for example, it may generate a critique where it’s supposed to generate a revision, or vice versa" (constitutional-ai, §"3.1 Method", p. 8). sl-cai-few-shot-examples fixes this not by changing the underlying instructions but by prepending consistently-formatted worked examples of the critique-then-revision pattern before the real conversation, so the model has a concrete template for which role belongs at which position in the context. Without this fix, the two-step pipeline that critique-and-revision depends on would periodically produce garbled output where a revision reads like a critique or vice versa.