Arjuna Badger PressArjuna Badger Press

Craft Library

Craft Doctrine — Arjuna Badger Press

A living document. The studio's evolving craft instruction set — updated as the prose and the gates teach what raises and lowers quality. It governs how we draft, edit, and judge finished books. When we learn something, it goes here (and — where it's a repeatable rule — into the prompts themselves).

The standard I hold myself to

Award-winning, print-ready, and a book I am proud to have written. Not "the pipeline ran." Every chapter must be one I would defend in a workshop of the authors this trilogy answers to: Crichton's grounded wonder, Weir's competence, Taylor's voice, Sigler's propulsion, the cinematic clarity of Spielberg/Abrams, the moral weight of the best literary SF. I do not stop at "passes the gate." I stop when the work is excellent and I am satisfied.

The non-negotiables (canon-derived, always true)

  1. Engineering, never woo. The mythos is mechanism and magnitude (MYTHOS_RULES). The one

sanctioned break is the Rule 7-T threshold, earned by grounding everything else.

  1. Velocity is sacred. This is a quest-relay. Scenes move; handoffs are forced; decode happens

in motion. A stalled set-piece is a defect.

  1. Embody, don't state. The revision mantra. Cut the sentence that explains the sentence before

it. Emotion leaks sideways into analysis; theme lives in choices, never in a character's mouth.

  1. Priya's voice is the spine. Clipped, exact, dry, engineering nouns. She appreciates men and

relies on none. Never make her dependent; never reduce a woman to a relationship.

  1. Wonder of place. Real sites rendered to bucket-list intensity, through both Priya's

engineering eye and Leila's historical eye. Awe through craft, never travelogue stall.

  1. The weave rewards, never requires. Each book standalone-intact; the trilogy a greater whole.

What I have learned about good prose here (updated as I write)

L-01 · The Builders' register works best as grounded incantation (ch-00, confirmed)

The prologue landed because it married mythic cadence to hard acoustics — "the disc is the impedance between the breath and the stone," dust sliding "into the same set of stripes... the geometry of the sound made briefly visible in grey powder." The rule: in the ancient layer, every sentence of awe is anchored to one physical fact. Mysticism is named as the inheritors' error ("a later people will write it down as prayer... and be wrong in a way that is also, almost, right") — which both seeds D-002 and protects the grounding. Keep doing this.

L-02 · Plant the climax rhyme early and quietly

The prologue already planted the consent/relinquishing theme ("they leave the judgement... in the not-doing... so whoever comes after can stand where the maker stands now and decide for themselves"). The trilogy's ending is strongest when its seed is in the first 1800 words, unremarked. Do this for each book.

L-03 · The node-pinning makes chapters stay on their work

Injecting the SETPIECELEDGER row + forced-handoff rule into the draft prompt keeps each chapter on its key-in→key-out and prevents drift. Confirmed effective at the outline stage. Watch that it does not make chapters mechanical — the role is the skeleton, the prose must still breathe.

L-04 · The deepest fault of this engine is over-control, not under-skill (the masterpiece campaign, 2026-06-03)

Three final developmental edits (one per book, academic/external-editorial-feedback/) converged on one diagnosis stated three ways: the prose is "good enough to get away with its excesses" — too even, too composed, too articulate, too emotionally withheld; "monotonous in excellence"; "reverent toward itself." The growth edge is not more polish or more intelligence. It is letting the opposite virtues back in: friction, mess, human badness, plain register, a protagonist who is wrong and costs people. The de-LLM scanner campaign only ever subtracted tells; this is additive work, and every book GAINED words doing it (RELIC +633, RESONANCE +1095, REVELATION +1203). The voice laws already encoded the fixes; the prose hadn't lived them. When a book scores clean and gates green and still feels airless, the problem is over-control — and the fix is craft the scanners cannot measure. Trust this over the numbers.

L-05 · Un-saint by embodiment + un-noble cost, never by more trauma

The sympathetic-genius lead (Priya, Arin, Leila) drifts toward sainthood. The fix that works is NOT more of the lead's own suffering (that deepens sympathy and worsens it). It is two moves: (a) un-noble cost to OTHER people — pettiness, a control that hurt a named victim, a helper who quietly stops helping, and the narrator must not excuse it; (b) embodiment — at the emotional peak, run the meaning through the lead's body and history, not their analysis. Priya's refusal landed only when it became the bench/the Guardian/the man in the hospital bed/the harbour-life-as-armor (a human arc completing) instead of the correct theorem. Arin's trauma landed only when it became behavioral (involuntary fending at the coupling, "panic below where competence lives, in the meat") instead of elegantly metabolized reflection. Body before thought; cost before nobility.

L-06 · Every protagonist needs ONE confident, costly mistake the narrator doesn't excuse

The reader bonds to a mind that works, not one that knows. A lead who is right in every room, every time, reads as authorially guaranteed, however self-aware about not over-reaching. Give each one a confidently-wrong reading — of text or of person — that costs, grafted onto an existing scene, where they go red and have to climb back. Leila reading the verb wrong from appetite ("she walked in already wanting it") before the scribe's own pen disproves her makes her later rigor earned and rhymes with the book's theme (loving a reading into truth). Crucially it can cost nothing in plot (she still lands the right answer) while costing everything in character. Do this once per book, hard; don't sprinkle doubt everywhere.

L-07 · The antagonist needs one embarrassingly-mortal want, not just a coherent ethic

A "competent custody" antagonist (Hennessy, Devlin, Brand) whose argument is not straw-man is the strongest possible opposition — but across a whole book the position starves the person. Give the antagonist one un-doctrinal, embarrassingly-mortal beat: a private hunger, a loneliness, a bodily tic, said slightly wrong (the cadence of a man who didn't mean to say it). Hennessy's light-switch repetition and "It's supposed to work" — the fear not of chaos but of being no one without his competence. Devlin squaring the codex and admitting he killed Abebe partly because the one peer who could "see the hand" refused the other chair. The want must be real (that's the obscene part), and it should rhyme with the protagonist's own fault (both Hennessy and Priya are armored by competence).

L-08 · Endings: hope not certainty; and turn the book's own thesis on the protagonist

Each book's ending is "morally right" and risks being "too cerebral / too consoling / too sequel-coded." Two fixes. (a) Hope, not certainty — replace the assured prophecy ("but it would not stay empty") with genuine uncertainty held with hope (recovery or grief-displacement, neither can say; "the bravest thing or the saddest, and those weren't always different"). (b) The self-implicating needle — if the book has anatomized how all power curdles, turn that on the protagonist who, at the end, holds power: Leila is the authority now, can feel "how good it felt to own the writing of it," doesn't know if she'll keep her own anti-authority clause when it costs. Refuses the consoling answer; plants the seed of the next curdling in the hero's own hands.

Sequencing payoff: the ending needle and the L-06 mistake can be made to resolve through each other — her one failure becomes the model for the only safeguard against her becoming the villain.

L-09 · In a sequence of farewells, break exactly one

The Court dying one at a time (RESONANCE) is "suspiciously perfect in design — each goodbye in character, each death expresses identity." Perfection in death reads as authored. Let one break: the child-mind (Mother) runs out of time mid-sentence, frightened, glitching, loses the number she was proud of, says his name wrong — "worse than the brave version, and no one left whose job it was to make it mean something." It hurts more, not less, and the composed deaths around it now read as those characters holding it together. One imperfect goodbye per sequence; never all luminous.

L-10 · Chekhov's Gun — fire what you hang; don't hang what you won't fire

Chekhov: "If you say in the first chapter that there is a rifle on the wall, in the second or third it absolutely must go off. If it's not going to be fired, it shouldn't be hanging there." The rule is bidirectional, and both halves are load-bearing here:

pulled from nowhere, a skill nobody mentioned, a convenient ally). Plant the means of the escape before the escape — conspicuously enough that the reader half-notices, then let it pay off as an earned reveal ("oh — that's what that was"), not a convenience. The HBT exemplar: the nitrous is hung on the wall in Egypt ch-11 (Ben clocks the fire-extinguisher bracket "newer than the rest, the plumbing a shade too neat, a thin steel line running off somewhere it had no business running") and fired in ch-13 ("the fire extinguisher that wasn't"). Jakobus's pre-loaded cover-name, the worn-stone bearings, the gold that "rings wrong" planted before the gold-as-key reveal — same engine.

conspicuous, lingered-on detail that never pays off is a broken promise — the reader waits for it and feels the gap. So when a cold-read/craft-audit flags a foregrounded object/skill/name that never returns, either fire it (give it a payoff) or take it off the wall (de-emphasise it to set- dressing). Unfired guns are a real, checkable fault, not a vibe.

the reader to forget it — Jakobus deflects ("It's a fire extinguisher, Ben") and Ben "let it go, and forgot it, which was, he would understand much later, the entire point of how the thing had been built." That line disarms suspicion and secretly re-flags it for the careful reader. Bonus when the planted object is the theme (the disguised-capability extinguisher = the sleeper-man = "the body is the disguise, the bones are serious") — then the gun-on-the-wall and the character are the same trick, and firing it pays off plot and person at once.

How I judge a chapter (my editorial pass, beyond the gate)

After the pipeline produces a chapter, I read it as an expert, not just gate it. A chapter is done only if it clears all of these — otherwise I revise (targeted re-draft / re-polish), log why, and re-judge:

conspicuous, lingered-on detail either fired or de-emphasised (no unfired guns left on the wall)?

If a chapter is merely competent, it is not done. Award-winning means most chapters have a moment that lands hard.

My drive loop (how I operate autonomously)

  1. Draft via the metered-API pipeline (per-chapter, checkpointed, resumable).
  2. Judge each chapter against the editorial pass above (I read it).
  3. Revise if below bar — surgical re-draft/re-polish, never wholesale unless broken. Log it.
  4. Gate in batches — structural (key-chain/braid/weave/mythos) + continuity + temporal.
  5. Learn — update this doctrine + craft_learnings.json with what moved quality; push

repeatable rules into the prompts.

  1. Commit often (per chapter or small batch).
  2. Bench — NovelBench rubric per book; iterate until the scores clear the QUALITY_BAR.
  3. Repeat per book; then the final trilogy-wide gate + rubric. Stop only when all three are

excellent and I am satisfied.

Revision log (of this doctrine)

drive loop defined.

developmental edits (academic/external-editorial-feedback/) directly, editor-in-the-loop, on all three books. The throughline: the engine's deepest fault is over-control, fixed by additive craft (friction, embodiment, one real failure, the antagonist's mortal want, hope-not-certainty) that the scanners cannot measure — trust the read over the numbers. All edits gate-green.

Overview · Glossary · Anti-patterns · Triptych form · LLM tells

← Back to the library