Tier 1 — Fable owns the whole deliverable (a voice/cadence gate is failing)
These are pure human-voice writing tasks. A quality gate is already failing on generic prose. Fable 5 is the creative-writing-strongest model; this is exactly its lane.
100Tarot voiceblocked
Rewrite the Etsy tarot / twin-flame response in Kate's actual cadence, then re-pass the Kate voice-similarity gate.
Why Fable: The last attempt scored 0.15 against a 0.65 floor. This is not a logic problem, it is a voice problem, matching a real reader's rhythm and diction. That is the single thing Fable is built to do better than Opus.
Next step: Feed Fable the 188 Kate transcripts + kate_combo_meanings.jsonl, have it draft the response in her cadence, re-run tarot_voice_v2_gate.py until score ≥ 0.65.
q-2026-06-21-f5928f
67Tarot voicequeued
De-fake the Chiron tarot-meanings corpus: strip AI-inserted bad grammar and restore Chiron's real speaking register.
Why Fable: The corpus is full of faked 'bad grammar' AI tells (ain't, 'what she don't', 'meat suit'). Restoring a believable natural register across hundreds of entries is a prose-voice rewrite, not a script edit. Fable holds a consistent human voice over long batches better.
Next step: Point Fable at the yt-chironintuition transcripts as the register anchor, rewrite the flagged entries, rebuild chiron-deck.json + /chiron-study.
q-2026-06-28-6eb676
Tier 2 — the ixi analysis-prose cluster (the 'believe-you-me' voice)
Seven BELIEVEYOUME v5 pages need their close-reading music-analysis PROSE rewritten to match a corrected authority note. The harmonic facts are fixed; what's needed is the analytical narrative voice. Fable writes the prose, the authority .md supplies the truth.
84Music analysis prosequeued
Align believeyoume-alligator-v5 prose to the authority md (5 sections; drop invented bridge/outro sections).
Why Fable: This is ixi-style close reading: 'voice supplies the third B3/C4, bitonal lead F# stays G-major over Bb bass.' Fable renders that as readable analytical prose while you hold it to the authority md's facts.
Next step: Hand Fable the authority md as ground truth, rewrite each section's prose only, one song per pass. Keep numbers/keys verbatim from the md.
q-2026-06-28-714e38
46Music analysis prosequeued
Align believeyoume-nottoomuch-v5 prose (bVII↔I sus swap; fix the wrong climbing-D-E-F read).
Why Fable: Same pattern: the analytical wording is wrong, the corrected reading is known. A prose-only rewrite against fixed facts is Fable's sweet spot.
Next step: Fable rewrites the descending C#-C-Bb-Ab section to the md, one pass.
q-2026-06-28-74f9a3
46Music analysis prosequeued
Align believeyoume-brokensat-v5 prose (Eb major flips to Eb minor; outro parks on D major, never resolves).
Why Fable: Replace a clumsy 'one D three jobs' framing with clear prose. Facts are locked, voice is the deliverable.
Next step: Fable rewrites to the authority md, one pass.
Why Fable: Close-reading prose rewrite against a corrected md. Fable.
Next step: Fable rewrites to the md, one pass.
q-2026-06-28-e95e72
Tier 3 — reader-voice generation + book prose
Generative interpretive prose in a named reader's voice, plus a book PDF regeneration. Fable authors the words; existing tooling handles extraction, gating, rendering, and publishing.
85Tarot voiceblocked
Tarot voice V3 Phase 2D: per-card-combo (78×78) deliverable bundler once v3_fused_rag clears the 0.92 gate.
Why Fable: The bundler is plumbing (PDF/MP3/video), but the CONTENT it bundles is per-pair interpretive prose. Fable should author the combo meanings that clear the 0.92 voice gate; the rest of the pipeline stays as-is.
Next step: Use Fable for the meaning-text generation stage only; keep XTTS + PDF-scroll render on the existing path.
Why Fable: Extraction is code, but the naming-decisiveness rescore rewards interpretation-verb prose over card-listing. Fable is better at producing (and judging) the interpretive-verb voice these readers use.
Next step: Keep the extractor in Python; use Fable for any voice-synthesis / rescore-authoring step.
Why Fable: Batch generation of per-pair interpretive prose in a reader's voice. Directly generative writing = Fable.
Next step: Route the 1C generation batch to Fable once the 1B gate clears. Same for quanyin / risingsoul / lightassefa.
q-2026-06-04-a9cb77
62Book prosequeued
Regenerate the BELIEVEYOUME Gumroad book PDF ('The Ground Never Settles') from the corrected v5 sheets.
Why Fable: A full book requires long-form connected prose. Fable holds a book voice across chapters. (Note: a Jackie-key dispute blocks republish, and republish needs your explicit OK.)
Next step: Fable drafts the reconciled book prose; resolve the Eb-vs-Ab key dispute before any Gumroad republish.
q-2026-06-28-7dee0b
46Book proseblocked
Read the ChatGPT conversation in-browser, produce a full critique-and-improvement, ship as HTML+md.
Why Fable: The deliverable is a written critique and improved rewrite, essay-shaped. Fable's prose edge helps the writing half; capture + deploy stay on Chrome + ship.py.
Next step: Capture via Chrome extension, hand the text to Fable for the critique/rewrite, ship.
q-2026-06-21-12ccc7
Not Fable — keep these on the current tools
Most open work is plumbing or signal-processing. Fable is a writer, not a deployer or a stem separator. Sending these to Fable would waste it.
Each of the 102 open ledger items was tagged by domain and scored: +creative signal (voice / cadence / prose / meaning / rewrite / reading / chapter / lyric), −plumbing signal (deploy / render / stem / MIDI / transcribe / synth / disk / curl). Items where a human-voice quality gate is the actual deliverable rose to the top. Fable 5 (claude-fable-5) is the creative-writing-strongest model, so it is routed to the writing stage only; extraction, gating, rendering, and deploy stay on their existing tools.
Source of truth: ~/.claude/shift_queue.jsonl (latest-by-id wins). Fable-fit is a routing recommendation for the prose-authoring stage of each task, not a claim that the whole task is done. Built 11 curated cards from the 12+ highest-fit items.