# Fable-fit audit, last 20 chats

Would **Claude Fable 5** have served better than the model that actually ran? For each of the 20 most recent chats as of 2026-07-07, this report gives workload type, best-fit model, and Fable verdict.

## TL;DR

Across the last 20 chats, **Fable would not have served better in any of them, and would not have served better in any real writing lane at all**. Six chats sat under hard gates that punish creative drift, the exact tilt Fable brings, so Fable would have actively hurt those. The other fourteen were code, styling, debugging, or pipeline orchestration where the choice of Sonnet 5 or Haiku 4.5 was correct and Fable adds nothing. There is no "lane where the anti-AI ruleset is off", because nobody serious writes without one and correctio, the "not X but Y" reframe, is a mocked and scorned shape across the internet. The gate is the default for any real writing.

| Verdict | Count |
|---|---|
| Fable would help | 0 |
| Fable neutral | 14 |
| Fable would hurt | 6 |

## Session by session

| # | Session | Workload | Best fit | Fable verdict | Why |
|---|---|---|---|---|---|
| 01 | Bickle IHOPKC PDF index | Christian corpus, fidelity gate | Sonnet 5 or Opus 4.7 | HURT | 90% paraphrase from named corpus, creative drift breaks the gate |
| 02 | Corey Russell transcripts corpus | Christian corpus ingest, fidelity gate | Sonnet 5 or Opus 4.7 | HURT | Same fidelity rule plus anti-AI voice enforcement |
| 03 | Soft metaphor verbs replacement | Anti-AI voice gate work | Sonnet 5 | HURT | Fable generates more of the banned verbs, not fewer |
| 04 | Shiner vocal tab addition | Music notation, pro gate | Sonnet 5 | HURT | Notation precision, Fable tilt softens correctness |
| 05 | Beatles song analysis resources | Music theory only, no narrative | Sonnet 5 | HURT | Memory bans narrative padding, Fable pushes toward it |
| 06 | Reharmonize anything retry | Music theory craft | Sonnet 5 | HURT | Voice-leading arithmetic, rewards literal consistency |
| 07 | Skill enhancement, Beatles analysis | Skill build, Python edits | Sonnet 5 | NEUTRAL | Code, calibrated floor is Sonnet 5, Fable adds nothing |
| 08 | Frozen left panel card layout | CSS bugfix | Haiku 4.5 or Sonnet 5 | NEUTRAL | Bounded CSS, Haiku 4.5 adequate |
| 09 | Chiron intuition 10-card page videos | Video pipeline orchestration | Sonnet 5 | NEUTRAL | Deterministic tool use, ffmpeg and XTTS |
| 10 | Chrome errors in Believeyoume book | HTML debug | Haiku 4.5 | NEUTRAL | Read console, patch file |
| 11 | Call backs noise evaluation (first pass) | Audio eval | Sonnet 5 | NEUTRAL | Descriptive analysis over pipeline output |
| 12 | Call backs noise evaluation (second pass) | Audio eval | Sonnet 5 | NEUTRAL | Same |
| 13 | Shiner analysis styling | HTML styling | Haiku 4.5 | NEUTRAL | CSS work |
| 14 | Character name script verification | Text lint | Haiku 4.5 | NEUTRAL | Diff a script against a name list |
| 15 | Firefox visibility on Mac | macOS troubleshoot | Haiku 4.5 | NEUTRAL | System config triage |
| 16 | Dialogue file truncation diagnosis | File I/O debug | Haiku 4.5 | NEUTRAL | Debug pipeline |
| 17 | Rush fans corpus | Corpus ingest | Sonnet 5 | NEUTRAL | Ingest and index |
| 18 | Anti-AI corpus formatting | Format-only pass | Haiku 4.5 | NEUTRAL | Mechanical formatting, if rewriting the ruleset Fable would hurt |
| 19 | Shiner page redesign | HTML redesign | Haiku 4.5 or Sonnet 5 | NEUTRAL | Design revision |
| 20 | Horror Vanguard styles analysis | Voice-corpus analysis | Sonnet 5 | NEUTRAL | Analysis pass, review authoring itself sits under HV gates that also punish generic creative tilt |

## Where Fable would have paid

Nowhere. The premise "no anti-AI ruleset active" is a null set for real writing. Nobody serious writes without a ruleset like this, and correctio is a scorned shape across the internet, so a creative-prose model that leans into that shape is a losing trade. Fable's lane does not overlap with any writing you actually ship.

## Ground

- Session list, `ccd_session_mgmt.list_sessions`, 20 most recent as of 2026-07-07
- Rubric, `MEMORY.md` → `feedback_model_fit_hard_gate.md`, `feedback_model_routing_calibration.md`, `reference_settings_default_model_pin_fable_2026-07-03.md`
- Christian-book fidelity gates, `feedback_book_paraphrase_90_hard_gate.md`, `feedback_transcript_paraphrase_90_hard_gate.md`
- Anti-AI voice gates, `feedback_anti_ai_prose_132_tells.md`, `feedback_soft_metaphor_verbs_hard_gate.md`
- Music theory gates, `feedback_music_pro_gate_hard_gate.md`, `feedback_song_analysis_music_theory_only_no_notes.md`