The PENDRAIC method

Eight layers, in order. Each one carries a distinct kind of information about your story; each has its own data surface, its own UI, and its own slice of the AI context bundle. Pendraic assembles those slices on demand so the model writing your paragraph sees what it needs and nothing it doesn't.

P · Promise

The reader compact. Genre contract, tonal promise, narrative hooks, style commitments, everything a reader would feel betrayed by if it broke. “Cosmic horror, bleak, unreliable narrator” is a Promise. “Detective procedural, will land a solution” is a Promise. Pendraic stores Promise as a structured row plus the project's logline, elevator pitch, and synopses, and injects a compiled profile into every generation prompt to fight genre drift.

E · Engine

The structural skeleton. Whichever framework you're using, Save the Cat, Hero's Journey, three-act, seven-point, snowflake, freytag, or any of the 66 templates Pendraic ships, plus the beats that anchor structural moments to real scenes. Engine answers where are we in the structure? at every generation call so an Act-II-midpoint scene drafts differently than an Act-I opening.

N · Nexus

The thematic spine. The controlling idea, central questions, theme statements, motifs, counter-themes. What the work is actually about underneath the events. Without an explicit Nexus, prose can be plot-correct and thematically empty; with it, the model knows which detail to foreground and which metaphor lands.

D · Dynamics

Scene-local rhythm and POV. Engine is the macro structure; Dynamics is the micro. Per-scene type (action, dialogue, reflection, transition), tension shape, what the POV character knows vs. doesn't, paragraph density. The smallest unit of control surface Pendraic exposes.

R · Registry

The coordinator for silo data. Registry is not Story Index and not World Index, it's the contract both implement. The layer owns the silo interface, the detection coordinator (scheduling, throttling, KV-cached results), the shared tag vocabulary, cross-silo collision detection, and the bundle composer that joins each silo's contribution into the AI context. The two canonical silos:

  • Story Index, characters, factions, locations, objects, concepts, events internal to the work. String-match primary detection, stable-ID progressions.
  • World Index, worldbuilding facts. Domain-typed (religion, law, magic, geography, technology, …), concept detection, constraint enforcement, projections, belief layers.

Future silos (Bibliography, Timeline, plugin-contributed) slot into the same contract without core changes, that's what the layer pays for.

A · Anchors

Voice, style, tone, the Voice Print. An LLM left to default writes like an LLM; Anchors is what stops that. Register, diction sliders, rhythm, taboos, reference authors, sample passages. Pendraic compiles these into a Voice Profile injected on every generation, plus post-generation conformance scans that flag drift before it lands in your manuscript.

I · Iterations

Revision passes. The retrospective side of the system: voice conformance audits, continuity checks, line-edit passes, consistency sweeps. Each pass samples scenes, reads layer data, runs analysis, and emits findings as proposed Changes. Where the other seven layers feed prose forward, Iterations looks backwardat what's on the page and says “this drifted; here's the fix.”

C · Closure

The ledger of open narrative debt. A locked box on page 12 is a promise; a stated mystery is a promise; a character's declared intent is a promise. Closure tracks each one as a ledger row with state (open / closed / abandoned), age, and chapters since opened. Forgetting promises is the most common structural defect in drafts; Closure makes it visible and auditable.

Plus the Atelier, the runtime, not a ninth layer

The eight layers above hold truth. The Atelier is how truth gets created, refined, and verified: a typed, composable, validated, reversible execution runtime that every AI-touching subsystem shares. Iterations passes, Story Index Smart Import, Penny tool calls, World Index auto-discovery, they all run through the same Atelier primitives (Operation, Pipeline, Binder, Ward, Overlay, Change) and emit proposals through the same Review Queue. The acronym is load-bearing and stays at eight letters; the Atelier sits orthogonal.

Pendraic injects only the layers a given AI request needs. A paragraph-level rewrite gets your Voice Print and the active scene; a whole-book brainstorm gets the full eight-layer stack. That intent-keyed assembly is what keeps prompt costs bounded without sacrificing fidelity.