Arjuna Badger PressArjuna Badger Press
The workshop — for authors & editors
What this is. The companion to The technology — written for published authors, working editors, and serious first-timers, not for people who need to be taught what a chapter is. It explains what Arjuna Badger Press actually offers: a finish line for books that already have a soul, in your voice — with an engine that measures and guards, never replaces your prose. The archer's eye. The badger's nerve. — arjunabadger.press
0. The one sentence
You bring the book — however messy — answer twenty sharp questions, click Go, walk away for a day, and come back to a manuscript that has been drafted, edited, continuity-checked, and merged to print-ready; your job is the final proofread.
That is not marketing copy. That is the shape the engine was built for.
1. This is not a beginner's toy
The free Craft Library exists so anyone can learn degree-level craft without an MFA. The workshop is something else.
It is for:
- The debut author who has never finished — but also for
- The mid-career author with a dozen markdown files, a blueprint, and no clean draft — and especially for
- The established author with published work, deep lore, a bloated draft, and a decade of notes
scattered across formats — the Patrick Rothfuss problem, the George R. R. Martin problem, the "I don't need ideas, I need finish" problem.
If you have already sold books, you do not need us to teach you structure. You need a studio that can ingest what you already have, lock a canon the pipeline can trust, run the editorial stack you cannot afford to hire full-time, and hand you back something you are willing to sign.
The engine does not talk down to you. It reads your material, asks what only you can decide, and does the labour you have been putting off since 2011.
2. What you bring (any shape, any mess)
Drop the whole pile. The workshop is designed to ingest:
| You have | Formats | |---|---| | Published or draft manuscripts | .txt · .md · .pdf · .docx · EPUB | | Personal notes & lore bibles | markdown, Word docs, scribbled notebooks | | Handwritten maps, timelines, character sheets | photographs (scanned or phone pictures) | | Partial chapters, alternate versions, cut scenes | all of the above, together |
You do not need to normalise it first. That is the importer's job — map your material onto a locked canon contract (characters, timeline, world rules, style, blueprint) that every downstream pass obeys.
Three front doors, one back end:
` NEWCOMER ─┐ spark of an idea, no draft yet MIDDLE ──┼──► LOCKED CANON ──► PIPELINE ──► PRINT-READY BOOK ROTHFUSS ─┘ published work + lore dump + bloated MS `
- Newcomer — the wizard: a guided Q&A that builds the canon from your answers (already exists
in the engine as engine/wizard.py; the web flow is the thin layer still to wrap).
- Middle — the importer: your messy markdown pile → canon contract + gap report + a short
follow-up wizard for the holes (engine/import_canon.py — built).
- Rothfuss — the ingester: your finished (or over-long) manuscript + notes → reverse-engineered
canon from the prose, continuity graph, finish report, surgical path to done (engine/ingestmanuscript.py, engine/reversecanon.py — built).
After the canon locks, every author gets the same engine — outline, draft, multi-role polish, continuity audit, hard graph gate, de-LLM scanners, merge to EPUB/PDF.
3. The flow you actually experience
Step 1 — Upload & ingest
You provide the books, the notes, the pictures. The system reads them, chunks scenes, extracts characters, places, timeline, and world rules into a story bible you can review.
Step 2 — The wizard (≈20 questions)
Not a hundred-page form. A focused interview — the forks only you can answer:
- Whose story is this, and what do they want vs. what do they need?
- What is the one sentence a reader would repeat?
- What must never change (names, history, voice laws)?
- Where does the draft bloat, sag, or break continuity?
The wizard proposes options at each fork; you choose. The model does not get to both propose and decide. That rule is why the output still sounds like a person.
Step 3 — Click Go
The pipeline runs checkpointed and resumable: draft chapter by chapter, polish with a six-role editorial stack, continuity audit, hard block on any graph violation, score against craft targets, merge. You do not babysit it.
Step 4 — Come back (~24 hours)
You return to a merged manuscript — structurally sound, continuity-clean, machine-tell scanned — ready for your proofread. Not "AI slop you must hide." A book that survived gates a human editorial team would charge six figures to approximate.
Your last job is authorial: read it, cut what only you would cut, sign it.
4. What we never do (the line that matters)
The engine measures and alarms. It does not replace your voice.
- We do not ghost-write Rothfuss's prose and hand him different sentences.
- We do show him where the bloat is, where the timeline breaks, where the key-chain stalls —
and run the finish pipeline on the cuts he approves.
- We do not let the model drive; a human seed plus surgical edits authors the soul.
- We do let the gatekeeper reject a flattened polish pass and fall back to the stronger draft.
That invariant is why established authors can trust it with work that already has an audience.
The de-LLM layer names and counts the sentence-level tells — "Not X. Y." reframes, em-dash chains, "the way…" similes, even cadence, AI vocabulary clusters — so your voice survives thinning, not replacement. See the full catalog with BAD→GOOD examples:
→ LLM tics & tells · Engineers: The technology · Craft Glossary
5. What is built vs. what is left
Honest inventory (2026):
Built today (the engine — this catalogue is the proof):
- Canon wizard (terminal Q&A) —
./book new - Canon importer + gap report
- Manuscript ingester + reverse-canon
- Full draft → polish → gate → merge pipeline
- StoryGraph continuity gate (hard block)
- de-LLM scanners + cold-read + craft audit
- Finish report for bloated manuscripts
Thin wrapper left (trivial relative to the above):
- Web upload UI — drag-and-drop your pile
- Web wizard — ~20 questions in the browser (port of existing terminal flow)
- "Click Go" project dashboard — progress over a ~24h run
The engine exists and works. The library you are reading is twenty finished novels worth of evidence. What remains is packaging — auth, upload, a browser on the wizard, a progress bar — not inventing the creative stack from scratch. That remaining work is trivial relative to what is already shipped.
6. For editors
If you edit professionally, think of the workshop as the editorial department you always wanted automated:
- Developmental — structure, spine, relay/key-chain, arc conformance (scored, not guessed)
- Line — rhythm, filter words, machine-tells, voice homogenization (named, counted, targeted)
- Continuity — geospatial-temporal graph gate; violations are precise, not vague "something
feels off"
- Copy-level correctness — verification gate for factual claims + both sides of contested history
The tools produce reports and blocks, not rewrites. You stay the judge; the engine is the microscope.
7. Write with us
The workshop is opening in tiers. If you have a manuscript in the drawer — or a series the world has been waiting on — get in touch.
Bring the pile. Answer twenty questions. Click Go.
Companion docs: The technology (engineering) · Craft Library (free craft reference) · Verification Gate (accuracy standard)