Compare the decision, not the slogan.

title: "AIP-C01 Decision Cheat Sheet" summary: "A compact decision framework for selecting, securing, evaluating, operating, and governing generative AI systems on AWS."

Production decision sequence

1. Define the task, users, failure cost, quality, safety, privacy, latency, cost, availability, and evidence requirements.
2. Establish data provenance, permission, classification, retention, and deletion.
3. Choose prompting, retrieval, customization, workflow, or bounded agency from measured need.
4. Enforce identity, authorization, tenancy, tool permissions, approval, and budgets outside the model.
5. Version every release input and evaluate it against a representative baseline.
6. Observe quality, safety, latency, cost, quota, and business outcomes together.
7. Roll back the complete release unit when a hard gate fails.

Use the matrices by hiding the recommendation and defending it from constraints. Alter one variable—freshness, action impact, latency, data sensitivity, human review, or regional support—and explain how the design changes.

Internal raw scores are recall and reasoning signals only. They do not predict AWS scaled scores, and the provisional question pools remain noindex until named technical review.