This page is a report on whether Fable would have helped in the last 20 chats. Fable is one of Anthropic's newer Claude models, tuned for creative prose. The answer here is no in 20 of 20, because the recent chats were music theory, source-fidelity Christian books, and the anti-AI voice gate. Those punish creative drift, so Fable would either match a stricter model or lose ground.
Would Claude Fable 5 have served better than the model that ran? For each recent chat, workload type, best-fit model, and Fable verdict.
The North Star gate requires 90% paraphrase from the named Bickle corpus. Fable's creative tilt would push wording off the source, breaking feedback_book_paraphrase_90_hard_gate.md.
Same 90% paraphrase rule, same reason. Also anti-AI voice enforcement on top. Fable is the wrong tilt on both axes.
WOULD HURTThe task was to strip "lands, sits, names, reaches for, leans on, hangs on, orbits, rests, rides" from prose. Fable's default register generates more of these, not fewer. A creative-prose model fighting the anti-AI gate is a losing trade.
WOULD HURTfeedback_music_pro_gate_hard_gate.md demands notation precision, correct voicings, correct chord functions. Fable's creative-prose tuning does not buy anything here and risks softening the notation logic.
The memory feedback_song_analysis_music_theory_only_no_notes.md forbids production trivia and narrative padding. Fable would push toward the very padding the gate bans.
Reharmonization is voice-leading arithmetic on top of a chart. It rewards a model that stays literal and consistent. Fable's tilt away from that costs more than it earns.
WOULD HURTCode work on a skill. Sonnet 5 is the calibrated floor. Fable does not improve code quality or tool-use accuracy.
NEUTRALBounded CSS problem. Haiku 4.5 is cheapest and adequate. Fable brings no advantage.
NEUTRALDeterministic tool orchestration, ffmpeg and XTTS wrangling. Any capable model works. Fable adds nothing.
NEUTRALRead the console, patch the file. Haiku 4.5 handles it.
NEUTRALDescriptive audio analysis over pipeline output. Sonnet 5 is the calibrated floor here.
NEUTRALSame reasoning as pass one. Fable brings no advantage.
NEUTRALCSS work. Haiku 4.5 covers it. Fable adds nothing.
NEUTRALDiff a script against a name list. Haiku 4.5 handles it.
NEUTRALSystem config triage. Haiku 4.5 is fine.
NEUTRALDebug pipeline. Haiku 4.5 covers it.
NEUTRALIngest and index. Sonnet 5 floor. Fable irrelevant.
NEUTRALMechanical formatting. Haiku 4.5 fine. Note, if the task shifts to rewriting anti-AI ruleset prose, Fable would hurt.
NEUTRALDesign revision. Fable adds nothing.
NEUTRALAnalysis, not review authoring. HV review authoring itself sits under reference_hv_reviewer_texture_gate_rebuild.md and reference_hv_spoken_voice_standard_gates_2026-07-04.md which punish generic creative tilt. Even the review authoring lane is not clean Fable territory. Analysis pass is neutral.