[Feature Request] Add model performance monitoring and transparent API behavior documentation

## Summary

When a newer model is released, older models (in this case Sonnet 5) appear to degrade sharply. I spent 4–5 hours stuck on a simple bug with Sonnet 5. Switching to Fable resolved it immediately.

If this is intentional routing or quality throttling, it is opaque and feels like a way to burn extra tokens. Please document expected model behavior and add a way to monitor quality regressions.

I have used Sonnet for more than six months across multiple accounts.

## What happened

1. I was debugging a simple issue in Claude Code using Sonnet 5.
2. The model looped, repeated itself, and failed to make progress for 4–5 hours.
3. I switched the model to Fable.
4. The same task completed successfully.

## Why this matters

- Large token spend with no useful output
- No signal that the model is underperforming
- No public documentation of how older models change after a new release
- Hard to trust Claude Code for production work if quality can drop without notice

## Requested outcome

- Transparent documentation of model versions, routing, and any known quality changes after a new release
- In-product monitoring or a warning when a session is looping / not making progress
- A way to pin a known-good model version so behavior does not silently change
- Confirmation that older models are not intentionally degraded

## Environment

- Platform: darwin
- Terminal: iTerm.app
- Version: 2.1.274
- Feedback ID: b0777965-efe8-403e-9a8b-e7b6e2c22a85

## Errors

```json
[]

No files, tests, or entry points are named. Start by separating the requested documentation, quality-regression monitoring, looping warning, and model pinning into scoped requirements. Done should define supported model behavior and version semantics plus a testable monitoring or warning outcome, but the issue does not identify implementation locations.

Written by the indexing model from the issue text.