Treat a disappointing first output as information about the prompt, not a verdict on the model. The discipline is diagnosing why the output fell short, then fixing that specific thing — not re-running the same prompt and hoping, and not just typing "make it better."
Three common root causes, roughly in order of how often they're the actual problem:
- Missing or ambiguous context — Claude didn't have the background it needed
- Under-specified instructions — the instruction was present but vague on one of the four specificity dials (§1)
- Absent or mismatched examples — no example was given, or the example didn't match the desired output
The loop: a weak output points back to a thin component in the structure from §1. "Too generic" often means the audience or desired outcome wasn't specified. "Wrong format" means the output-format component was missing or unclear. "Not what I meant" often traces back to a missing example. Diagnosing which component is thin — and fixing that component specifically — is the iteration skill being tested here, as distinct from simply retrying.