Self-Consistencyadvanced
Self-Consistency
A three-part framework that asks the LLM to reason through a problem multiple ways and commit to the most consistent conclusion. Improves robustness on judgement calls.
The components
- 1Problem framing
A clear statement of the question to answer.
- 2Diverse reasoning paths
Instruction to produce N independent attempts that reach a conclusion via different routes.
- 3Synthesis
Instruction to compare the conclusions and commit to the most consistent answer with stated confidence.
Worked example
Rough prompt
Should we pivot our SaaS to focus only on the healthcare vertical?
Super-prompt produced by Self-Consistency
[Problem framing] Recommend whether the company should pivot away from its current generalist SaaS positioning to a healthcare-only vertical focus.
[Diverse reasoning paths] Generate three independent reasoning paths to a recommendation. Each path must approach the question through a different lens: (Path A) financial — unit economics, CAC payback, market size. (Path B) competitive — defensibility, incumbent strength, time-to-leadership. (Path C) organisational — team's existing expertise, founder fit, cost of becoming credible. Each path produces an independent recommendation with a one-sentence rationale.
[Synthesis] Compare the three recommendations. If two or more align, commit to that recommendation. If all three diverge, explain why and recommend the one whose underlying logic you assess as most load-bearing. Close with a stated confidence ("high / medium / low") and the single thing that would change your mind.