Diagnosing a Character

A character reads as thin because a load-bearing piece is absent, and the fastest way to find which piece is to run a fixed checklist against the character rather than reread the scene, because the missing piece is usually not on the page being stared at. The felt symptom (“something’s off, I can’t place it”) is almost never local; it is a structural hole two pages over — a duplicated arc type, an unbuilt web edge, a payoff with no plant — and the checklist surfaces exactly those.

The eleven checks

Run each against the character and mark it present, thin, or absent, with the on-page evidence. Each check has an owner page that carries the full treatment.

1. Wound / Ghost — a concrete backstory injury that installed the character’s false belief. The Wound and the Lie
2. Lie — the false belief about self or world the character holds at page one, shown as behavior. Same page.
3. Want vs Need — a nameable external goal that stands in for, and fights, the internal truth; if winning the Want would also grant the Need, the arc has no gap. The Change Arc
4. One defining contradiction, shown before named — a single interior tension that distorts a costly choice on the page. A trait pile is not a contradiction. Round Characters and the Telling Detail
5. Distinct voice — diction, rhythm, tics, and the topics the character refuses to name, so a line is assignable before its tag. Character Voice
6. Arc type, and whether it duplicates the cast’s dominant type — the single check that most often explains “well-built but thin,” because a competent arc reads flat when it rhymes with five others. Arc Types
7. Signature object / motif — a concrete thing that carries the interiority the scene has no room to state. Motif and Symbol
8. Plant → payoff chain — every arc-closing beat paid for by a quiet plant the reader already lived through. Companion Arcs and Party Banter
9. A home lesson for the crisis — the pivotal beat owns a scene, not an assertion in the bible and not a slot shared three ways. Spotlight Rotation
10. Web edges — built, revealing relationships to more than one other character, each doing relationship-work and revelation-work at once. Foils and Pairings · The Character Web
11. Submerged backstory — the iceberg’s hidden mass fully worked out, so a withhold reads as loaded rather than hollow. The Iceberg and World as Pressure

Two of these are cross-cast checks, not local ones: arc-type duplication (6) is read against the whole ensemble, and a web edge (10) can only be graded by looking at the other end. This is why a single character can pass every local check and still read as thin — the fault lives in the relationships between checks, not inside one.

How to read the marks

Thin wants deepening; absent wants building. An arc-type mark that reads “strong, but a duplicate” is the trap case: the individual character is fine and the cast is monotonous, so the fix is a re-cast elsewhere, not more texture here. A cluster of thins around checks 4, 7, and 11 (contradiction, object, backstory) is a texture problem, cheap to fix with a few telling details. An absent on check 3 (want equals need) or check 6 (no distinct arc) is a structural problem that rewrites the arc.

Worked example — 星野遙香

Running the checklist on the Tsumugu super-fan, whose author felt her “missing something I can’t place”:

Check	Mark	Evidence
Wound / Ghost	thin	The dreaded Osaka OL future is present; why it is assigned (the family, the specific pressure) is unwritten.
Lie	present	”The assigned life can be outrun.”
Want vs Need	present	Want: escape (visa hacks, another year). Need: author how she meets the path.
Defining contradiction	thin	Bright-over-dread is real but carried by one beat (the camera).
Distinct voice	present	Bright, quick, genki, Kansai; idiolect specifics unwritten.
Arc type (duplication)	strong	Disillusionment — the cast’s only clean non-positive arc. Her one clear strength.
Signature object	present	The camera that hoards proof of a freedom she fears losing.
Plant → payoff	thin	B2L14 seed → B3L08 crack, but the summons-home trigger is marked “optional/TBD.”
Home lesson	thin	B3L08 shares her mask-crack with 阿迪’s failure and 金多恩’s art crisis.
Web edges	absent-ish	Only 星野↔林薇 is built; the 阿迪 ship is thematic, never staged.
Submerged backstory	absent	The “missing context” the author senses.
Second defining challenge	absent	Every other lead has two axes; she has one.

The pattern the marks expose: her arc is the strongest in the cast, and everything around the arc is thin or absent. The felt thinness is not the arc failing; it is a single strong arc standing alone with no second axis, no built relationships, and no submerged mass under it. That is a diagnosis the author could not reach by rereading her scenes, because nothing is wrong inside them.

The handoff prompt

The checklist ports to a fresh agent as a prompt. The template, with the adversarial clause that keeps the agent from rubber-stamping prior research:

Diagnose and deepen <CHARACTER> in my <PROJECT>.
READ: <canon files> and the craft lens at wiki/Story Craft/ and <any prior diagnosis>.
FELT GAPS (symptoms, not conclusions): <what feels off>.
CALL THE RESEARCH INTO QUESTION: where has a flattering label (e.g. "the only X arc")
  excused leaving the character under-built? Does the prior fix address the felt gaps
  or paper over them? Resolve, don't dodge, any tension between what I want and what
  the research warns against.
PRODUCE (ranked options, I decide): (1) the eleven-check diagnosis with evidence;
  (2) 3-4 candidate additions on axes not yet occupied, each with its risk and its
  one seeding plant; (3) 6-10 shown-not-told texture beats, vocab-legal at their book;
  (4) 3-4 designed cast interactions that reveal both parties; (5) the submerged
  backstory, written even though most never ships; (6) a revised spine.
CONSTRAINTS: <canon locks, the vocabulary ceiling, the writing standard>.
Everything invented is a labeled proposal to approve, never a decision.

A worked instance is at tsumugu-core/docs/companion/craft/HANDOFF-xingye-diagnosis.md.

The case against

The checklist diagnoses; it does not write. A character that passes all eleven can still be lifeless if the prose enacting the beats is flat, and no structural audit catches a dead sentence. It also biases toward leads: a peripheral character (a running-gag cat, a background adult) is supposed to fail most checks, and running the full list on one manufactures work that flattens a deliberately flat character. Match the depth of the audit to the character’s rank — leads get eleven checks, flat support gets one memorable note and a graceful exit.

Price

A full eleven-check pass on a lead is twenty to thirty minutes of honest reading, and it earns its cost only on a character the author already senses is off, or on a lead before it locks. Running it on every character in a large cast on a schedule is busywork; the audit is a diagnostic reached for on a felt symptom, not a standing gate.

Quit signals

• Every check comes back present and the character still reads thin: the fault is in the prose, not the structure. Stop auditing and rewrite the scene.
• The audit generates a fix list longer than the character’s page count: the character is over-scoped for its rank. Cut checks to match how much room the character actually gets.
• Two leads return near-identical mark patterns: the shared fault is ensemble-level (arc-type monotony, one theme unspread), and belongs to The Character Web, not to either character.

Checkable expectations

• A character the author calls “thin, can’t place why” should return at least one absent among the eleven within one pass. If every check is present, the diagnosis was wrong about the symptom and the problem is prose-level.
• The arc-type check should be readable only against the full cast: grading it on the character alone means the check was run wrong.
• After building the absents and deepening the thins, a second pass should move at least those marks up a grade; if it does not, the additions were decoration, not structure.

Related

• The Character Web — the ensemble-level version of this audit, when two characters share a fault.
• Arc Types — check 6, the one that most often explains “well-built but thin.”
• Round Characters and the Telling Detail — checks 4 and 7, the cheap texture fixes.
• Story Craft — the wing this checklist indexes.