Sectioning or Voting: Parallelizing a Contract Review

A team parallelizes contract review two different ways and asks you to name which sub-variant of parallelization each is:

Design A: Three LLM calls run concurrently on the same contract — one checking for correctness of the extracted term dates, one checking for unusual indemnification language, and one checking for missing signature blocks. A fourth call merges the three sets of findings into one report.

Design B: The same "check for unusual indemnification language" task is run three times independently on the same contract at nonzero temperature, and the majority verdict (flagged / not flagged) is taken as the final answer.

1. Name the sub-variant for each design and justify with the distinguishing question.
2. For Design B, at what point would running more copies (say, 20 instead of 3) stop being worth it, and why?
3. A stakeholder proposes merging A and B into "one big parallel pattern" for simplicity. Explain why keeping them conceptually separate still matters even if they run inside the same pipeline.