Model Fallback + Robustness

The goal is not “never fail”, but “fail in a way you can reason about”.

Entry points:

• Model failover
• Providers: failover

What to keep explicit

• error normalization into a stable reason/status/code
• candidate generation with dedupe + overrides
• abort semantics (don’t fallback on user abort)
• attempts trail for logs/alerts/postmortems