Most worldbuilding tips for novelists assume the world is finished before the draft begins. Mine assume the opposite: you are forty pages in, a scene has just asked a question you have not answered, and the choice in front of you is to stop and build, to guess and move on, or to find a third way. This guide is about the third way.
I wrote a fantasy novel, Medalion, and later built Lokee, a tool for keeping a story's world straight, so I have thought about this from both sides. The method here does not depend on the tool. The first two thirds work with a notebook and a document; the last third shows how the same habits look when the vault does some of the remembering for you.
The running example is The Ashen Circle, the demo story in Lokee: Rowan Ashe, a scout turned informant, moving across the rooftops of Duskhollow while Magistrate Ophelia Crane closes in. Beside it I'll use an invented coastal town, Thornwater, where the sea rises a hand's width a year and every citizen owes a year of labour on the wall. The guide on starting worldbuilding built Thornwater from nothing; this one keeps it in step with a draft.
Why building as you write beats building first
A scene is the best worldbuilding prompt there is. It asks exactly one question at a time, in the order the book needs the answers, and it tells you how much of the answer matters, because the scene either continues or it does not.
Take chapter 1 of The Ashen Circle. Rowan drops into the alley behind the Gilded Lantern "before the patrol bell finished its second ring". That line raises a question: how does the curfew work? The scene needs only this much: there is a bell, it rings more than once, and being in the street after it is a problem. It does not need the Watch's duty roster, the bell's founding or the penalty schedule. Those may come up in chapter 3, when the magistrate's reports mention a dye-shop owner's brother "seen twice near the eastern wall after curfew". Now the question is sharper, "what counts as being seen?", and the answer is sharper too.
Building first would have produced the roster, the founding and the penalties in week one, and probably a curfew that chapter 3 then had to fight. Building as you write produces three sentences, each decided when the book could say what it needed.
The loop is small. A scene needs an answer. You decide the smallest rule that settles it. You note the rule somewhere you will find it again. You write on. The dead end on the other side of the diagram is "build everything first", and it is a dead end for a specific reason: nothing in it tells you when you are done.
The smallest rule that settles the scene
When a scene stops because the world has not answered, the temptation is to answer thoroughly. Resist it. Ask instead: what is the smallest rule that lets this scene continue and does not box in the scenes after it?
A good smallest rule has three properties. It is a single sentence. It is stated in terms a character could know. And it leaves the mechanism open.
Three examples, two from the demo and one from Thornwater:
- Chapter 2 needs Iseult to have been funding the Circle for a while without the Duke noticing. Smallest rule: the Duke inspects the cellar twice a year and never goes past the racks. Not how the passage was built, not who else knows. Two inspections a year is enough to let her plan around them.
- Chapter 3 needs Ophelia to connect Rowan to Mira without a confession. Smallest rule: the Watch keeps written reports and the magistrate reads them side by side. That is all. It also, usefully, makes paper a danger for both sides for the rest of the book.
- In Thornwater, Elin needs to find that the wall ledger does not balance. Smallest rule: every year of service is entered twice, once by the Warden who assigned it and once by the one who signed it off. Now a discrepancy is a thing one clerk can see, and the question "who faked the second entry?" is the plot's engine. I have not decided who. I do not have to yet.
What the smallest rule avoids is the extra clause. "The Duke inspects the cellar twice a year, at the spring and autumn tastings, with his steward Harl, who keeps the keys." Harl is now a character, the tastings are events, and the keys are a plot device you will have to honour. If the scene did not need them, do not write them down as decided. If you like them, put them on the not-yet list.
Worldbuilding tips for novelists who keep falling into research
Research is the respectable form of not writing. It feels like work, it produces notes, and it can go on indefinitely. Three habits keep it in its place.
Write the question before you open the book. "How did medieval cities enforce a curfew?" is a question with an answer, and you will recognise the answer when you find it. "Read about medieval cities" is not a question and has no stopping point. If you cannot phrase what the scene needs as a question, the scene does not need research; it needs a decision.
Give the question a budget. Twenty minutes, one chapter of one book, three search results. When the budget is spent, decide with what you have. The rule you decide can be revised later if the research would have changed it, and nine times in ten it would not have.
Keep a "decided without research" list. This is the list that makes the budget bearable. Write "curfew bell: rings twice, second ring is the deadline; not researched" and the anxiety that you are getting something wrong turns into a line you can check when the draft is done. Most of these lines you will never need to check, because the reader accepts the world the characters accept.
A related trap is the plausibility spiral, where one detail demands another. The sea rises a hand a year, so where does the water come from, so what is happening to the climate, so what about the other coasts. Stop at the first "so". The reader of Thornwater needs the wall, the labour and the rising. The cause can stay below the waterline for the whole book, and if it ever surfaces it will surface as a reveal, which is worth more than a footnote.
Exposition discipline: keep the iceberg under water
The iceberg from the starting guide applies twice as hard while drafting, because now you have notes and the notes want to be used. The test for whether a world detail belongs in the text is not "is it interesting?" It is one of these three:
- Does it change what a character does in this scene? The second bell changes when Rowan moves. It belongs.
- Does the reader need it to follow what is happening? The reader needs to know the Circle uses aliases, or "Kestrel" is just a name. It belongs, once.
- Does it pay off later? The Duke's twice-yearly inspection pays off when a third inspection is announced. It belongs, planted lightly.
A detail that passes none of the three stays in the notes. That is not waste. It is what let you write the scene with confidence, and confidence shows on the page as characters who treat their world as ordinary. Old Bram does not explain the tavern's neutrality; he says "She's already here. Corner table. Didn't give a name — didn't need to." The world is in the "didn't need to".
Two habits help. When you catch yourself writing a paragraph that begins with how something works, cut it and see whether the scene still reads; it usually does, and the paragraph was for you. And let exposition arrive as friction: a character bumps into a rule, pays for it or routes around it. Rules that cost something teach the reader faster than rules that are described.
When the plot changes, the world has to change with it
This is the part building-first never prepares you for. In chapter 14 you realise the book is better if Aldric already suspects Rowan in chapter 1. That is a plot change, and it has world consequences: the Watch's information is now better than you had said, Ophelia's reports in chapter 3 read differently, and the "old friend who doesn't yet know" line in Aldric's description is wrong.
The method is the same whether your notes are on paper or in a vault:
- Name the rule that changed, in one sentence, old and new. "Old: Aldric does not know what Rowan has become. New: Aldric suspects from chapter 1 and says nothing."
- List the scenes the old rule touched. Search the draft for the character, the place or the phrase. Chapter 1 (the doubled patrol, "Aldric's doing"), chapter 3 (what Ophelia has not told him), Aldric's own notes.
- Decide which way the fix goes. Sometimes the plot change is not worth the ripple, and that is a legitimate answer. Decide it looking at the list, not at the one scene that inspired the change.
- Fix every listed scene in one pass. Not as you happen to pass them in revision, because you will not pass all of them.
- Record the decision with its date, so that in chapter 30 you know which version of Aldric is current and why.
In Thornwater, a plot change might be: the ledger discrepancy is not fraud but a dead Warden's unfinished work. The rule "every year is entered twice" survives; what changes is who is guilty, which touches every scene where Elin suspects someone. The list is what makes the rewrite finite.
If you keep the record in Lokee, step 5 has two different shapes, and the distinction matters. If Aldric's suspicion is a canon edit, a correction to what was always true, edit his card and let the Revisions page log it. If it is something that becomes true at a moment in the story, use Record a change… beside the field at the world point where it happens, so that viewing chapter 1 still shows what was true then. The recording changes guide is about telling those two apart, and the three kinds of when guide explains why "when did the reader learn it?" is a third question again.
A weekly habit that keeps the world and the draft in step
The world drifts away from the draft one unrecorded decision at a time. The fix is not more discipline per sentence; it is a short regular pass. Twenty minutes once a week is enough, and it looks like this:
- Collect the week's open questions. The brackets in the text, the lines on the "decided without research" list, the thought you had in the shower. Put them in one place.
- Answer the ones the next chapter needs. Smallest rule each. Defer the rest, explicitly.
- Record the decisions made in passing. Read the week's pages looking only for sentences that state a world fact for the first time. Mira's shop is in the lower city; the East Gate has a watch house; the Undercroft is under the old quarter. Each one goes into the notes, attached to the entity it describes.
- Check the rule the plot bends around. Reread it. Did anything this week bend it by accident? In The Ashen Circle, that is "no one in the crown's service knows a real name". If a scene this week had a guard say "Nightingale", you have either a plot point or a mistake.
- Prune. Anything you decided and then wrote around, mark as dropped rather than deleting it, so you do not re-decide it next month.
Twenty minutes a week is cheaper than the three days it takes to reconcile a 300-page draft with a bible that stopped being true in chapter 9.
How drafting becomes worldbuilding in Lokee
Everything above can live in a notebook. In Lokee the same habits are mostly built into where you already are, which is the point of using it. Here is how each one maps, with the exact labels.
Deciding in passing, without leaving the chapter. In the Writer, type @ and start a name. If the entity exists, the list offers it; if not, the last item reads Create "name" as new entity. The mention becomes a link to the entity's card, and the entity exists from that moment, with nothing on it but a name and a type. That is the right amount of worldbuilding for a first appearance. You can also select a phrase and press Link entity on the selection toolbar to attach it to an existing entity or create a new one from the selection.
Catching the names you did not link. Passive mention suggestions are on by default: an unlinked entity name or alias is lightly underlined while you write, and a click offers Link to that entity or Ignore. Exact and alias matches only, and nothing links by itself. This is why aliases are worth adding early: "Kestrel" on Sera's card is what lets the Writer notice her in a sentence that never says "Sera". The switch lives under Settings → Writing as Suggest entity mentions while writing.
Recording the week's decisions from the text. Select the passage where a scene stated several world facts at once, the cellar paragraph in chapter 2 for instance, and press Extract selection; Extract this chapter does the same for a whole chapter. The AI proposes entities and relationships and every proposal goes through a review step before anything is created, so the pass is a reviewed import, not an automatic one. The draft-to-bible guide walks through the review screen in detail. Extraction uses Lokee credits in cloud vaults; in a local vault you can use your own key or a local model instead, with the cost shown before it runs.
Capturing a decision away from the desk. The microphone button on every vault page opens Voice Capture, with a Voice tab and a Text tab. Say or type "the Duke inspects the cellar twice a year and never goes past the racks", press Extract entities, and review what it proposes with Review proposals; or keep it as prose and choose Save as: a Vault note or a Board sticky note. The credit cost is shown before transcription or extraction runs. This is where the shower thought goes, instead of into whichever notebook is nearest.
Keeping the rule where you will see it. A rule is prose, and Lokee has no special object for it. Put it in a note under Notes (+ New note), and use + Link to… to attach it to the entities, chapters and maps it governs, so it shows up on their pages and, with Show notes on graph turned on in Settings, on the Graph beside them. The "decided without research" list and the not-yet list are two more notes. The notes-organising guide covers folders and tags for when there are many of them.
The weekly check, done for you in part. Consistency Check in the sidebar reads the whole vault and lists findings under Problems, Worth a look and Just so you know: someone in two places at once, a relationship that ends before it begins, a date that disagrees with the order you set. It flags and never fixes, and the continuity check guide explains each kind of finding. It cannot see a rule stated in prose, so the "did I bend the central rule?" question stays on your weekly list. It is also hidden for reviewers and on the phone.
What Lokee does not do here. It does not decide the smallest rule, it does not know when research has answered your question, and it will not tell you whether a detail belongs in the text or in the notes. Extraction finds names and relationships, not consequences: it can propose "Duke Cassian Varnhold, Parent Of, Lady Iseult Varnhold" from chapter 2, but the rule about the cellar inspections is yours to phrase and record. The vault removes the remembering and the searching, which were never the hard part, so that more of your attention is left for the deciding, which is.
Common mistakes mid-draft, and the fixes
Answering the scene's question with a system. The scene needed a bell; you wrote a city ordinance. Fix: one sentence, the character's point of view, mechanism left open.
Guessing and not marking it. You give the Watch House a dungeon in chapter 5 because the scene wanted one, and forget it is now canon. Fix: any new world fact in the text gets a bracket or a line in the week's list. In Lokee, an @ mention or a captured sentence is that mark.
Researching the wrong question. Three hours on medieval dye-making for a scene whose real question was whether Mira's shop is on a street the Watch patrols. Fix: write the question first.
Letting the notes write the prose. A chapter opens with the history of the Undercroft because you had it. Fix: the three tests. Cut anything that passes none.
Fixing the plot change one scene at a time. You change Aldric in chapter 1 and chapter 14, and chapter 3 still has the old version. Fix: list before edit, one pass.
Overwriting the old rule. You replace the sentence in your notes and three weeks later cannot remember why chapter 3 reads strangely. Fix: keep the old rule, dated, marked as superseded.
Skipping the weekly pass. Fix: shrink it rather than skip it. Five minutes of "what did I decide this week?" keeps the gap from compounding.
Before your next writing session
A short checklist, in the order it helps:
- Read the one rule the plot bends around.
- Look at the open-questions list; answer only what the next scene needs.
- Open the chapter with the entities it uses beside it; in Lokee, clicking a mention opens the card.
- Write. Mark new world facts as you make them; do not stop to record them.
- At the end, spend five minutes moving the marks into the notes.
Where to go next
If you have not yet made the first five decisions, the starting guide is the place to begin; it ends where this one starts. When the notes outgrow a document, how to organise worldbuilding notes turns scattered files into one system, and the story bible guide designs the types and fields that keep a vault useful at three hundred entries. For an honest tour of what the tool does and leaves to you, read how Lokee helps with worldbuilding. The help pages on the Writer, Capture and Notes list every control named above. If you would rather try the loop than read about it, the desktop app at /download opens a local vault with no account.