bug: selectable model catalog is stale — Claude Fable 5.1 (claude-fable-5-1) not selectable anywhere; tier labels + alias help text behind the current lineup

[vybe]

Summary

The selectable Claude-model catalog is behind Anthropic's current lineup. Claude Fable 5.1 (claude-fable-5-1) — Anthropic's most capable widely released model, the successor to Fable 5 in the same tier at the same price — cannot be selected anywhere in Trinity: not in the operator model picker, not as the admin fleet default, not as a per-agent public-channel override (the server-validated allow-list rejects it with 422), and not in the Workspace composer dropdown.

Every list is mutually consistent and CI is green, because they all derive from one source — but that source has not moved since Fable 5.1 shipped. This is exactly the source-vs-reality staleness that #2086 documents as the one remaining human control ("when Anthropic ships a selectable Claude model, add one entry and re-run codegen").

Context

• Update and actualize the available LLM model list across the app (add Opus 4.8, retire EOL models) #1080 (Opus 4.8) and bug: Fable 5 / Sonnet 5 missing from platform default-model dropdown and PUBLIC_CHANNEL_MODELS whitelist (drifted from ModelSelector presets) #1660 (Fable 5 / Sonnet 5) were the last two catalog refreshes; refactor: centralize the selectable model catalog — one source of truth for three hand-synced lists #2086 then centralized the catalog so a refresh is a single-file edit + codegen. Nothing has added a model since — the last touch (2026-09-09, ent#403) only added the Workspace dimension.
• Current Anthropic lineup, verified against the claude-api reference rather than stated from memory (the catalog's own rule): Fable 5.1 (claude-fable-5-1, current — most capable), Fable 5 (claude-fable-5, still served, now prior-gen in its tier), Opus 5, Opus 4.8 / 4.7 / 4.6, Sonnet 5, Sonnet 4.6, Haiku 4.5. Claude Mythos 5.1 is access-gated (Project Glasswing) and does not belong in a general catalog.
• Two adjacent copies of model-generation wording have gone stale with it: the picker's tier notes (the "(latest)" markers on Opus 5 / Fable 5 / Sonnet 5 — Fable 5 is no longer the latest in its tier) and the agent-server chat router's alias help text, which still says the opus alias means Opus 4.8.

Acceptance Criteria

• Claude Fable 5.1 (claude-fable-5-1) is selectable end-to-end: operator model picker, admin fleet-default dropdown, per-agent public-channel override (PUT /api/agents/{name}/public-channel-model returns 200, not 422), and the Workspace composer.
• Tier notes / "(latest)" markers reflect the current lineup — Fable 5.1 is the latest in its tier; Fable 5 is no longer labelled as such.
• The agent-server alias help text (opus / sonnet / haiku / fable) names the generation each alias currently resolves to.
• Model ids are the exact undated ids (claude-fable-5-1 — never a date-suffixed variant).
• The generated frontend mirror and the catalog parity guard (tests/unit/test_2086_model_catalog_parity.py) are in sync; CI green.
• docs/memory/feature-flows/model-selection.md preset table + changelog updated.
• No default changes. The platform default (PLATFORM_DEFAULT_MODEL_VALUE) and the recommended marker stay where they are — that is a cost/latency policy call, explicitly out of scope for this issue.

Technical Notes

• Single source of truth: src/backend/services/model_catalog.py; the frontend mirror src/frontend/src/constants/modelCatalog.js is generated by scripts/gen_model_catalog.py (never hand-edited). The file's own docstring describes the whole maintenance surface.
• Alias help text lives in docker/base-image/agent_server/routers/chat.py (base-image change → note the image rebuild).