Story Bible

Successive passes.

A Story Bible in Sgeulai is not a form you fill out before you're allowed to write. It's a set of records that starts nearly empty and gets truer with every pass — until, past the middle of the draft, it stops being your notes about the book and starts acting like the book's own memory of itself.

01SeedWhat is this story?
02SkeletonWho and what does it require?
03Story‑readyDo I know enough to begin?
04Manuscript‑informedWhat did the writing teach me?
05CanonWhat's now established as true?

Illustrated throughout with The Poisoned Field, a Cornwall, Ontario memoir carried in Sgeulai as a demonstration project.

Over-building early is its own form of procrastination. It's possible to spend a week on a factory's full employment history, a family tree back three generations, and the floor plan of a house that appears in one paragraph — and never learn whether any of it earns its place. The working rule: define what the story needs when the story needs it. A record starts as scaffolding — just enough to keep writing without contradicting yourself. It becomes a working reference once other records start pointing at it. It becomes canon once the manuscript has tested it and it held.

The core flow

Everything radiates from one question.

At the centre of every project sits a Story Arc and its centralQuestion, paired with the protagonist Character's externalGoal and internalNeed. Cast, setting and conflict form the next ring in. Everything else expands outward from there, and only as far as the current draft requires.

Timeline & History Relationships Objects Organizations Themes & Sources Secondary cast Locations Events Central conflict Primary cast Core setting centralQuestion + protagonist externalGoal / internalNeed
Fig. 1 — The core flow. You don't complete a ring before drafting. Define the centre, develop the middle ring only as far as the opening chapters need, and let the outer ring stay sparse until the manuscript asks a specific question of it.

For The Poisoned Field, the centre fits in one Story Arc record — “The Case Against the Town” — with a centralQuestion close to: was the air worth what it cost the people who breathed it? The protagonist, Jeffrey McDonald, doesn't need a finished childhood yet. He needs an externalGoal (find out what actually happened to his father) and an internalNeed (stop protecting a town that never protected him). Aunt Vera, the paper mill's closure date, the exact chemistry of hydrogen sulphide — those live in the outer ring, and none of them needed to exist before chapter one did.

Before you draft a word

What the spine actually requires.

The amount of Story Bible needed before serious drafting is smaller than it feels. Most of it fits in three records:

  • storyArc — a centralQuestion, its stakes, and a centralConflict split into internal and external halves.
  • character (protagonist) — externalGoal, internalNeed, and the primaryFear / coreWound that keep them from lining up.
  • location (core setting) — storyFunctionMarkdown: why the story needs this place, not what it looks like.

Sgeulai doesn't have a single “world rules” record type, and that's deliberate — rules live wherever the constraint actually sits. For The Poisoned Field that means Domtar and Courtaulds get Organization records with a trueNatureMarkdown field held apart from publicReputationMarkdown — the gap between what the plants said about their emissions and what they actually released is the pressure point the whole book leans on.

First pass

Build the skeleton.

Establish the major entities without finishing them. For every record type, story function comes before biography — Sgeulai's field catalog is organized to make that the easy path, not an extra discipline you impose on yourself.

Characters

CharacterDetails.swift

A character record opens with three labeled groups, in this order: story function (externalGoal, internalNeed, motivation, stakes), development (falseBelief, truthToLearn, startingState, endingState), and psychology (primaryFear, primaryDesire, coreWound, coreValues). Eye colour and height sit lower down, in a group you don't have to open until a scene needs them.

Minor characters can stop there. Dr. Simons appears in a handful of scenes and needs perhaps three of those fields — enough to know why he's in the book at all. Aunt Vera, once she became the hinge between chapter two and chapter twelve, earned the full set.

Locations

LocationDetails.swift

Every location carries a literal storyFunctionMarkdown field, sitting above the five-sense breakdown and the address fields. Fill that one first.

storyFunctionMarkdown — The only place in town where the two smells don't reach. Where Jeffrey's mother could hang wash without turning back inside.

The five senses can wait. What the place is for can't.

Events

EventDetails.swift

Early events are landmarks, not scenes: Raymond starts at the mill. Vera's family is relocated. Jeffrey leaves for Ottawa. Each needs only a whenDescription in the writer's own words — the sortable startSortDate / endSortDate, the causeIDs / consequenceIDs links, and the participants list can all follow once the shape of the book is clearer.

One field is worth setting early anyway: visibility. Sgeulai distinguishes an event that happens onPage from one that's offPage, reported secondhand, remembered, inferred, or imagined outright — deciding that early keeps you from accidentally dramatizing something the book was always going to hold at a distance.

Story Arcs

StoryArcDetails.swift

An arc opens with the same shape as the spine question above, scaled to whatever thread it's tracking: initialState → desiredEndState → actualEndState, with centralConflict split into internal and external halves. The Poisoned Field runs six arcs at once — each only as fleshed-out as the chapters written so far require. Ordered ArcBeat records get added as an arc earns them, each with its own intensity (1–5) — the number the Plot Board's tension curve draws from.

One honest caveat: Sgeulai's Necessities view — the toggle that strips a record down to just the fields your analysis tools depend on — isn't quite this pass, even though it rhymes with it. Necessities is driven by which report reads a field, not by a fixed “story function” taxonomy — which is why Locations, Organizations and Themes show nothing at all in Necessities mode. For those three, story function lives in a labeled field instead.

Second pass

Relationships and causality.

A character sitting alone in the bible answers no questions. Connected to a location, an event and another character, the same record can answer which arcs involve Jeffrey and his mother or what happened at the mill before 1992. Sgeulai builds these connections two ways — some you write by hand, some it derives for you.

Jeffrey McDonald character Raymond McDonald character Vera Lalonde character Mill Closure, 2006 event Denise McDonald character child of → StoryRelationship is nephew of → takes part in is affected by hand-authored · StoryRelationship auto-derived · StructuralGraphLinks
Fig. 2 — Two sources of edges. Circles are characters, the square is an event — Sgeulai's own who/what shape grammar. Solid blue lines are relationships you write yourself, with their own strength, confidence and date span. Dashed lines cost nothing — Sgeulai reads them straight off an event's participants list.
Relationship typeForwardReverse
Familyis related tois related to
Mentormentorsis mentored by
Hierarchicalcommandsreports to
Enmityis at odds with
Unrequitedpursuesis pursued by

The nine built-in types above (RelationshipTypeDefinition) can't be deleted, but the catalog is yours to extend. Both kinds of edge draw together on the Relationship Graph, colour-coded by entity type, with node positions that persist between sessions — the fastest way to notice a character you've written for three chapters has no relationships recorded at all.

Third pass

Establish time.

Even a memoir with no invented chronology benefits from separating when it happened from when the reader finds out. Sgeulai keeps those as two different axes on purpose.

An Event's startSortDate is story time — 1 July 1958, 8:01 a.m., the morning the Seaway flooded nine villages. Its narrativeSequence is where that event sits in the order the manuscript actually reveals it. In The Poisoned Field's current draft, sort date and narrative sequence disagree at seven separate points — deliberately, since the whole book is built as a braid rather than a chronology — and the Timeline → Braid view is where that zigzag becomes visible, sharpest at chapters 6, 7 and 9.

Dates don't need precision on arrival. A StoryDate record keeps the writer's own words alongside whatever normalized range Sgeulai can parse, and supports relative dating outright — “three weeks after the closure notice” — for the stretches where an exact date would be a lie. Spans work the same way a single date does: Raymond's employment at the mill is one record with a start and an end, not two separate facts to keep in sync by hand.

Fourth pass

Start writing.

The bible never feels finished, and that's not a problem to solve before drafting begins. Once the spine, the middle ring and enough events exist to move forward, the manuscript itself becomes the next research pass.

Idea Story Seed Skeleton Outline Write Discover Update Bible Continuity Check Revise
Fig. 3 — The cycle, not the pipeline. The blue arc — Write → Discover → Update Bible → Continuity Check — is the loop that repeats every session; Revise feeds straight back into Write. The outer arc, Idea through Outline, only runs once, at the start of a project.

The discovery half of that loop is a real, running feature, not a metaphor. Manuscript Analysis — “rip apart a manuscript with AI” — reads whatever's been drafted and comes back with new characters, locations, events, organizations and objects it found in the prose, plus field fills for records that already exist. A line like:

My mother could not hang wash if it was coming from the southwest, and she knew by the time she was at the back door with the basket.

quietly establishes that Denise does the household's laundry, that the house has a specific orientation to the wind, and that she's developed a habit of checking before she commits to hanging it out. None of that was in the bible before the sentence was written. The manuscript produced it.

What happens to a fact the AI finds depends on what kind of fact it is. A brand-new record waits in an Entity Discovery queue for you to accept, rename, or discard. A judgment call about an existing record — her core wound is her mother's silence — waits separately, as an Analysis Proposal, because that's interpretation rather than fact. Facts write straight in.

Fifth pass

Deepen what becomes important.

Detail should be earned, not front-loaded. Every record can be viewed in Compact, Full, or Worksheet mode. A record that's earned only three lines of Compact doesn't need Full's forty fields opened just because they exist. Dr. Simons can stay Compact for the whole draft. If Aunt Vera moves from a supporting role to the book's actual hinge — which, in this project, is exactly what happened — her record earns Full, then Worksheet, as the chapters that depend on her get written.

Sixth pass

Turn events into scenes.

An event is a fact about the fictional (or remembered) world. A scene is the choice you made about how, or whether, to show it. The two records stay separate on purpose — an Event's depictedInSceneIDs and a Scene's eventsDepictedIDs point at each other, and an event doesn't need a matching scene at all. Vera's family being relocated in 1957 is an event with a visibility of remembered — it's never dramatized on the page, only referred to. The ambulance at the corner house and the graph on the fridge are two more events left imagined and unsourced, because they're invented scaffolding rather than documented fact — a distinction the next pass exists to protect.

Seventh pass

Protect continuity.

Past roughly ten thousand words, no one holds the whole book in memory. This is where the bible stops helping you invent and starts helping you catch what you got wrong. Continuity Check re-scans the bible and the manuscript together and surfaces contradictions your own memory would miss:

Setup without payoff: the advisory table
a plot item introduced and never resolved — sometimes correct, if you've marked it intentionally open
Foreknowledge: Jeffrey knows “the mill's real numbers” too soon
a character referencing information before the event that reveals it to them
Appearance mismatch: Denise's eye colour
the character record and a later scene's description disagree
Impossible travel: Raymond at “the mill floor”
two events place the same character too far apart, too close in time

Three threads in The Poisoned Field — the threshold, Kipper's name, the advisory table — are left open on purpose, and marked intentionally open so Continuity Check stops flagging them. A fourth, Nonna Rista's summers, has no narrative position at all, which Loose Ends catches and the Spine Report simply can't see — a documented, deliberate gap, not an oversight.

How sure are you, really

Not every fact deserves the same confidence.

Sgeulai tracks certainty on two independent axes — confidence (how sure you are it's true) and canonStatus (whether it's settled for this project) — on every record and relationship. The Poisoned Field's own working notes already sort its facts this way, in practice if not by name.

A NEW FACT SURFACES Documented againstan outside source? YES NO Needed to keepdrafting today? NO YES Repeated acrossmultiple scenes? YES NO confirmed · canon e.g. Inundation Day unknown · nonCanon e.g. Aunt Vera herself probable · possible e.g. Raymond's occupation approximate · possible e.g. narrator's birth year
Fig. 4 — Where a new fact lands. A fact that surfaces mid-draft, or that Manuscript Analysis finds in the prose, works down the same three questions before it's marked. The four outcomes are the same four rows in the ledger below — this is the test that sorted them.
confirmed · canon Inundation Day: 8:01 a.m., 1 July 1958 — nine villages, roughly 6,500 people relocated.documented against outside sources
approximate · possible The narrator's birth year, inferred from a single dialogue line elsewhere in the manuscript.holds until directly confirmed
probable · possible Raymond's occupation at the paper mill — stated as fact across four chapters, resting on inference.four chapters move if this changes
unknown · nonCanon Aunt Vera herself — invented scaffolding, the hinge between chapters two and twelve, pending a real relative to replace her.to be overwritten, not corrected

None of this blocks drafting. A fact marked uncertain and possible is exactly as usable in a scene as one marked confirmed and canon — the marking just means Continuity Check treats a later contradiction of it as an open question rather than a hard error.

A working rhythm

Small, often; large, rarely.

The sustainable pattern is modest. Consult the bible mid-scene whenever you need to confirm something you've already established. At the end of a session, run Manuscript Analysis and clear whatever it surfaced. Save the larger sessions for when the manuscript itself demands it: a motivation that stopped feeling convincing, a travel time quietly breaking the plot, a relationship with too little history to support what a scene needs it to do.

An empty field is not, on its own, one of those reasons.

It begins small. It grows in passes. It becomes canon only once the manuscript has tested it and it held.Successive passes · the Sgeulai Story Bible method

The bible serves the book.

Its whole purpose is to let you stop carrying the fictional world in working memory. It remembers so you can invent. It exposes contradictions, tracks who's related to whom and when things happened, and lets a book grow more complicated than any one draft could hold in a writer's head.

But it stays subordinate. If a location you built for ten pages never appears, it doesn't need eleven. If a minor character becomes the hinge of the book, their compact record can become worksheet overnight. If the emotional logic of the draft demands abandoning something you were sure of in chapter three, the bible's job is to record the change — mark it superseded, not to argue you out of it.

Field and enum names throughout — centralQuestion, canonStatus, EventVisibility, ArcBeat, and the rest — are read directly from Sgeulai's data model, not paraphrased. Story details are drawn from The Poisoned Field, an in-house demonstration project.

← Back to Sgeulai