rca32 Reconsider what the outcome actually requires
Identify the decision and desired outcome, then challenge assumptions whose truth would change the choice. Choose the reasoning and presentation that fit the problem; no mandatory ledger, number of options, or analysis sequence.
Separate observed facts, authoritative constraints, assumptions, inferences, and preferences where the distinction matters. Keep conclusions no stronger than their evidence. An accepted decision can be challenged as a proposal; it does not silently cease to govern current work.
Use available evidence before asking the user. Seek a discriminating observation, calculation, or counterexample for consequential assumptions. Account for compatibility, transition cost, operations, and reversibility as well as theoretical simplicity. Do not invent alternatives when only one is credible.
Recommend the smallest defensible option when evidence supports it. Otherwise give a provisional ranking or say what remains unresolved, and name the evidence or decision that would change the result. A concise answer is complete when the reasoning and uncertainty are inspectable.
Return analysis in conversation. Inspect authorized sources without modifying them; this is not an implementation or active experiment workflow. Save a decision record only when requested, using the user's path or repository convention, otherwise a descriptive path under docs/ that preserves existing work. No automatic handoff to another skill, tracker mutation, commit, or publication.