[Bug] Fable model fallback behavior causes poor user experience in tasks

Bug Description
I'm genuinely disappointed, but this Fable fallback mechanism is frankly ridiculous. It's not just a minor inconvenience — it's an aesthetic and functional offense to Claude Code users who rely on the model for serious, specialized work.

Let me be direct: I don't train LLMs. I don't work with biology, cybersecurity, natural sciences, or any of the domains listed in that support article. My work is quantitative research in finance — algorithmic trading strategies, statistical modeling, portfolio optimization, risk management, time-series analysis, Monte Carlo simulations, and reinforcement learning for medium-frequency trading systems. This is a completely different field, with its own vocabulary, mathematical foundations, and methodological rigor.

The problem is that the fallback system seems to have a shallow, keyword-based trigger that misclassifies "quantitative research" as one of those unrelated topics. When I'm discussing concepts like Sharpe ratios, drawdown analysis, Extreme Value Theory, delta-neutral strategies, or D4PG architectures, the system apparently gets confused and decides I need to be handed off to Fable — a model that clearly isn't equipped for this level of technical depth.

This is deeply frustrating because:
It breaks workflow continuity. I'm in the middle of a complex reasoning chain about trading algorithms, and suddenly the model switches to something that can't keep up with the technical context.

It shows a fundamental misunderstanding of the domain. Quantitative finance is not "science" in the way the fallback system seems to interpret it. It's applied mathematics, statistics, and computer science — and it requires a model that can handle rigorous formal reasoning, not just general knowledge.

It makes Claude Code unreliable for professional use. If I can't trust that the model will stay consistent throughout a conversation about my actual work, I have to constantly second-guess whether I'm getting the best possible response or if the system has silently downgraded me.

The sad truth is that this is pushing me — and likely many others — toward alternatives like Codex CLI. I don't want to leave Claude Code, but when the product actively undermines my ability to use it effectively, I have to consider options that don't have these arbitrary and poorly-implemented fallbacks.

To the team: please, reconsider how this fallback is triggered. Quantitative research in finance is a legitimate, demanding use case that deserves proper support — not a one-size-fits-all switch to a model that can't handle the specificity of the work.

Environment Info

• Platform: linux ubuntu
• Terminal: cursor
• Version: 2.1.201
• Feedback ID: 3db8baf4-fba3-4899-b1c1-08f76d37ed74

Errors

[Fable 5's safeguards flagged this message. This sometimes happens with safe, normal conversations. Switched to Opus 4.8. Send feedback with /feedback or learn more]