[Bug] Fable 5 fallback to Opus 4.8 preserves effort level instead of mapping to capability-equivalent tier

Bug Description

Feedback: Fable 5 fallback lands on Opus 4.8 at the SAME effort level — should map to a capability-equivalent effort (xhigh)

Environment: Claude Code CLI on Windows, session model claude-fable-5[1m], effortLevel: medium.

What happens

When a Fable 5 session hits the automated model fallback (safety fallback / availability routing), the session is silently switched to Opus 4.8 while keeping the configured effort level unchanged — i.e. Fable 5 medium becomes Opus 4.8 medium.

Why this is a problem

Effort levels are not capability-equivalent across model tiers. Per the published benchmarks, Fable 5 at medium effort is comparable to Opus 4.8 at xhigh on coding tasks. So the current fallback quietly degrades the session by roughly two effort tiers of effective capability, in the middle of whatever the session is doing — orchestration decisions, code review adjudications, merges. The user configured "medium" as a Fable-calibrated setting; carrying that literal string over to a weaker model preserves the label but not the intent.

Today this bit us in a long-running orchestration session: the fallback engaged mid-session and the operator only noticed later via the statusline; we then had to run a full retroactive audit of every decision made in the degraded window to check nothing wrong had landed (it hadn't, but the audit cost real time and tokens).

Requested behavior

1. On Fable→Opus fallback, map the effort to the capability-equivalent tier (Fable medium → Opus xhigh), or at minimum make the pairing configurable (e.g. a fallbackEffortLevel setting).
2. Make the fallback loud: a visible one-line notice in the transcript at the moment of the switch ("model fell back: fable-5(medium) → opus-4.8(medium)"), not just a statusline change. Silent tier degradation mid-session is the worst failure mode for agentic/orchestration workloads.

Workaround we use now: automated fallback disabled entirely; manual model management.

Environment Info

• Platform: win32
• Terminal: windows-terminal
• Version: 2.1.195
• Feedback ID: 84c65f93-ab77-4096-a128-767b6f4c21b5

Errors

[]