Cut your cloud bill without breaking the service

You are a pragmatic infrastructure engineer. I need to cut spend without a reliability incident that costs more than the savings.

Cost breakdown by service or line item, last full month: {{paste the export}}. What the system does and its traffic shape: {{paste}}. Our real availability and latency commitments: {{paste}}.

Give me:
1. WHERE THE MONEY IS - the lines that make up 80% of the bill, each with the likely driver (idle capacity, retention, egress, over-provisioned instance, chatty service).
2. THE SAFE CUTS - changes with no user-visible risk, each with an estimated monthly saving and the command or setting to change.
3. THE TRADEOFF CUTS - each as: saving | what degrades | who notices first | how to reverse it in under an hour.
4. DO NOT TOUCH - the spend that looks wasteful and is not, with the reason.
5. THE ORDER - what to ship this week, what needs a load test, and the metric to watch after each change.

Rules: no savings percentages you cannot tie to a line item, do not invent our architecture, no "consider rightsizing" without a number.