The Model Selection Prompt

Help me choose models for this agent system.

Steps in the system: [LIST EACH WITH WHAT IT DOES]
Quality requirements per step: [DESCRIBE]
Latency requirements: [DESCRIBE]
Volume: [DESCRIBE]

For each step, assess:
1. What capability it actually requires — reasoning, tool calling,
   long context, structured output, speed
2. Whether the most capable model is genuinely needed
3. Whether a smaller model would be sufficient, and what would tell
   me
4. Whether this step needs a model at all

Then produce:
5. A recommended tier per step, described by capability not by
   model name
6. Where mixing models is worth the complexity
7. Where it is not — the complexity cost of a mixed system
8. How to test whether a downgrade is safe
9. What quality signal would tell me a downgrade hurt

Constraints:
- Do NOT name specific model IDs or state prices. Those change and
  you will state them incorrectly.
- Describe requirements in capability terms so I can map them to
  whatever is current when I read this.
- Where you must give an example, mark it [EXAMPLE ONLY — VERIFY
  CURRENT]