---
name: learning-loop
description: "Use verified MemQ outcomes to improve future task decisions through bounded recall and authorized reflection."
---

# Learn from verified outcomes

Apply this workflow when an observed result can improve future work. Verify project/account scope before protected memory operations. Use only currently exposed tools and their schemas.

1. Identify the observation and its evidence: a test result, source revision, service response, or directly observed outcome. Separate hypotheses from measured results.
2. Recall a small number of relevant prior procedures or failures. Repeated memories are not independent evidence.
3. When authorized, save one concise outcome with its source, conditions, limitations, and a reusable next-step rule.
4. Use reflection or consolidation only when it has a specific benefit and its write scope is authorized. These operations may mutate data and consume quota; they are not mandatory session-close tasks.
5. Test the proposed lesson on new work before presenting it as established. Retain contradictory results and uncertainty.

This workflow improves retained context and procedures; it does not promise automatic model-weight training or guaranteed improvement. Never reinforce a claim merely by repeatedly saving it. Do not publish private lessons into shared or public memory without authorization.