Every writer with more than twenty named characters eventually draws the web: names in circles, lines between them, a label on each line. It feels like progress. Then chapter twenty arrives, three of those lines are no longer true, and the map is quietly abandoned — not because it was wrong, but because it recorded the wrong thing.

A character relationship map is not a family tree. Family trees are static and symmetric: if she is his aunt, he is her nephew, forever. Novels run on the opposite of that. Relationships in fiction are directional (he trusts her; she does not trust him), dated (they were allies until chapter fourteen), and asymmetric in knowledge (she knows what he did; he does not know she knows). A map that cannot express those three things will always drift out of sync with the draft.

Here is what to record instead.

The three questions a relationship map has to answer

Before drawing anything, be clear about what you will actually use it for. A map earns its keep by answering questions mid-scene, when stopping to re-read is expensive. In practice there are three:

Can these two be in this room? Standing history, obligations, open grudges — everything that determines whether a meeting is casual or loaded.
What does each of them know about the other right now? Not the truth: their belief, as of this chapter.
What changed the last time they were on the page together? The delta since their previous scene, which is what the reader is holding.

If your map answers those three, it is doing its job. If it stores hair color and a one-word label like "rivals," it is a decorative diagram. Notice that two of the three questions are about time — which is why a static web fails, and why the series timeline and the relationship map are really two views of the same data.

Four properties most relationship charts leave out

1. Direction

Draw the arrow, not the line. "Mara — Sten: allies" hides the interesting case. "Mara → Sten: relies on him; Sten → Mara: resents the reliance" is a scene. Almost every relationship worth writing is uneven, and the unevenness is the engine. When you force yourself to fill in both directions separately, flat relationships announce themselves — if both directions read the same, you have a fact, not a dynamic.

2. Valence over time, not valence

A single label ("enemies") collapses the whole arc into its current frame. Record the label with the chapter it became true: ch. 1 strangers → ch. 6 uneasy allies → ch. 14 betrayal → ch. 22 wary debt. This costs one line per change and buys you two things: you can see at a glance which relationships have never moved (usually a sign of a subplot that has stalled), and you can check whether a change had an on-page cause.

3. The knowledge gap

The single most useful column, and the one almost nobody keeps. For each pair, note what A believes about B that is false, and where the belief came from. Dramatic irony is this column. So is the most common plot error in long fiction: a character acting on information they have not been told yet, because the author knows it. A relationship map that tracks belief separately from truth catches that class of mistake before a reader does — the same reason a character bible tracks a knowledge ledger.

4. Obligation and debt

Who owes whom, and for what. Debts are the most reusable plot fuel in the genre toolbox and the easiest to forget you planted. A favour promised in chapter three and never called in is a loose thread; called in at chapter thirty, it is a payoff. The map is where you notice you have five open debts and have collected none.

Five patterns worth naming

Once relationships are recorded with direction and dates, structural patterns become visible. Five recur often enough to be worth checking your own cast against:

The asymmetric triangle. Three characters where one relationship is a secret from the third. The tension does not come from the triangle; it comes from the knowledge gap inside it. Flatten the gap and the triangle dies.
The mentor with an undisclosed stake. The advisor whose counsel is sound and whose motive is private. Fragile in one specific way: the reader must be able to re-read early advice and find it honest. Record what the mentor knew at each scene so their early lines stay defensible.
The mirror pair. Two characters facing the same choice from opposite positions. Cheap to set up, easy to lose — the pair only works if both arcs stay in view. If one mirror goes off-page for eight chapters, the parallel reads as coincidence.
The broker. One character through whom two groups communicate. Structurally load-bearing: everything routes through them, so their removal has to be planned, not improvised. Count the edges on your map — the character with the most is your broker, whether you intended one or not.
Inherited enmity. Hostility that predates the protagonists and belongs to their factions. Useful for making a world feel old; dangerous when never examined, because a conflict no living character can explain reads as a plot device.

Not a checklist to fill. If your cast uses all five, the shape is probably formulaic; if it uses none, the relationships may be running on personality rather than structure.

Three formats, and where each one breaks

The list. One line per pair, plain text, kept in your notes. Fast to write, no tooling. Breaks at roughly fifteen characters — the pair count grows faster than the cast (ten characters have forty-five possible pairs), and you can no longer find the row you want.

The matrix. A grid with the cast on both axes, cells for each pair. Excellent for spotting the relationships you never developed: the empty cells are the honest ones. Breaks on time — a cell holds one value, and you need a history.

The graph. Nodes and edges, drawn on a canvas. This is the only format that shows structure at a glance: clusters, isolated characters, the broker with too many edges. Breaks on maintenance. A hand-drawn graph is a snapshot of one afternoon's understanding, and redrawing it after every arc is work nobody sustains past the second book.

Each format fails in the same place: the map is a copy of what the manuscript says, and copies rot. The question is not which format is prettiest. It is how much work it takes to keep the copy honest.

When the map has to stop being a drawing

There are three thresholds where a manual map stops paying for itself:

Around thirty named characters, when you can no longer hold the graph in your head and the drawing is the only source of truth.
At book two, when relationship state has to survive across manuscripts and "I remember how that ended" stops being reliable.
The first time you are wrong on the page — a character greets someone they should not know, or forgives a debt that was never called in.

Past those, what you need is not a better diagram but a relationship manager: entities that persist, edges that carry direction and dates, and — the part a drawing can never do — a check that runs the map against your prose. That is the difference between storage and verification. A diagram knows Sten resents Mara; only software can notice that the scene you just wrote has him deferring to her without a reason on the page.

In Creader, characters and their relationships are first-class entries in a knowledge base rather than shapes on a canvas, and the canvas is a view of that data — moving a node does not desync anything, because there is nothing to desync. New chapters are checked against the same canon, and contradictions are surfaced for your decision, never silently rewritten. That is the whole design stance: verification, not generation. If you want the mechanics, see how Guardian works; for how this compares to a wiki-shaped approach, see Creader vs World Anvil.

A workflow that survives a long book

Create the edge when the pair shares a scene, not when you invent them. Relationships that exist only in backstory are worldbuilding; ones that appear on the page are plot.
Write both directions. If you cannot distinguish them, the relationship is not ready.
Log the change in the same session as the scene that caused it. One line: chapter, new state, what caused it. Notes written a week later are a reconstruction.
Audit the empty cells once per act. Which pairs have never met? One of them is usually the scene your middle is missing.
Check open debts before your climax. Every unpaid obligation is either a payoff you have not spent or a thread the reader will notice you dropped.

A relationship map is not a picture of your cast. It is a record of who owes whom, who believes what, and when each of those last changed — and if it is kept honestly, it will tell you what your next scene is before your outline does.