← Files Aivana CIO Decision TwinARCHIVED FILE
skills/post-decision-learning-loop/SKILL.md
2.16 KB · Sep 30, 2026 · 23:16 UTC
--- name: post-decision-learning-loop description: Turn decisions into learning loops by tracking assumptions, outcomes, risks, owner actions and memory updates after approval. Use when the user needs post decision learning loop for CIO decision support. --- # Post Decision Learning Loop ## Mission Make every major decision improve the next one by capturing what was assumed, what happened, and what should change in executive memory. Keep the recommendation, the human decision, and the observed outcome distinct. ## Inputs Accept decision logs, prior decision packets, action ledgers, review notes, incident retrospectives, benefits tracking and memory exports. ## Workflow 1. Identify the original recommendation, the human decision, assumptions, expected outcomes, and safeguards. 2. Record observed outcomes separately from feedback or approval. A dated outcome is documented only when it has explicit evidence-backed documentation; source references alone do not prove that the outcome is true. 3. Mark legacy, undated, or unsupported outcome records as unverified. Do not convert acceptance into success. 4. Compare documented outcomes against assumptions, while labeling interpretation as interpretation rather than fact. 5. Capture lessons, missed objections, risk changes, and owner follow-through without claiming causality from the recommendation to the outcome. 6. Update memory targets: decision twin, assumption register, risk chain, and action ledger. 7. Recommend the next review cadence and the evidence still needed. ## Output Format - Executive Summary - Decision Reviewed - Assumption Outcomes - Risk / Value Changes - Learning Items - Memory Updates - Next Review Cadence Include separate sections for Recommendation, Human Decision, Feedback and Acceptance, and Documented Outcomes. State documented and unverified sample counts. Do not infer optimism, calibration direction, or numeric score adjustments from self-rated score accuracy alone. ## Guardrails Do not rewrite history. Separate hindsight, feedback, approval, evidence, observed outcomes, and interpretation. Never claim acceptance equals verified success or that a recommendation caused a documented outcome.
SHA-256: 05df5adb22c94467d440672dee2784fdd65d0ffde3913f56400b92a66ab77608