Skip to content

Audit reasoning

Check whether a conclusion is sound before trusting it, by sorting evidence from inference, reconstructing the inferential climb, and reviewing through separated lenses.

graph LR
  nthinkevidencevsinferencesort["Evidence vs Inference Sort"]
  nthinkladderofinferencecheck["Ladder of Inference Check"]
  nthinkparallelperspectivesreview["Parallel Perspectives Review"]
  nthinkevidencevsinferencesort --> nthinkladderofinferencecheck --> nthinkparallelperspectivesreview

Each step passes a compressed artifact to the next. The numbered list below is the same chain in text.

  1. think-evidence-vs-inference-sort
  2. think-ladder-of-inference-check
  3. think-parallel-perspectives-review

Job: check whether a conclusion or recommendation is actually sound before trusting it.

Use when: a conclusion feels confident but consequential, or you want to audit the reasoning behind a recommendation (the agent’s own or someone else’s). This recipe operationalizes the library’s epistemic positioning.

  1. think-evidence-vs-inference-sort (skills/think-evidence-vs-inference-sort/SKILL.md)
    • Sort the claims behind the conclusion into evidence, inference, and assumption; flag the uncited.
    • Carry forward: the load-bearing unknowns (unsupported claims doing the heavy lifting).
  2. think-ladder-of-inference-check (skills/think-ladder-of-inference-check/SKILL.md)
    • Reconstruct how the conclusion was reached; flag the riskiest leap; test an alternative interpretation.
    • Carry forward: the riskiest rung and the alternative interpretation.
  3. think-parallel-perspectives-review (skills/think-parallel-perspectives-review/SKILL.md)
    • Review the conclusion through separated lenses to surface what a single viewpoint missed.
    • Carry forward: the synthesis and the central tension.

A reasoning audit: which claims are actually supported, where the inferential leaps and alternative readings are, and a rounded verdict on whether the conclusion can be trusted or needs more before it is acted on.

Carry only the flagged items between steps (unknowns -> riskiest rung -> synthesis), not the full ledgers and traces.

Thinking Framework Skills v0.3.0 · 38 frameworks