You have a draft. The whole arc. Beginning, middle, end. People who read it tell you it's good. You agree, mostly. The sentences hold. The characters are alive on the page. The plot resolves. And yet — and this is the part that has been gnawing at you for some weeks — it isn't quite working. There is a softness around the climax. The ending is technically resolved but the resolution doesn't quite land. The middle has the right shape but its weight is off by some fraction you cannot quite name.
This is the almost-works draft. It is the most common failure mode in fiction. It is also, importantly, the most diagnosable.
The reason it is the most diagnosable is that an almost-works draft has been completed, and a completed draft is rich evidence. Everything is on the page. Nothing has to be inferred. The diagnostic does not have to ask "what does this story want to be?" It can ask the much simpler question: "where does the chain break?"
The chain is the dependency structure of the seven layers. In a coherent novel, each layer's condition is explained by the layer beneath it. Voice is explained by Structure: every POV decision is governed by the structure's informational architecture. Structure is explained by Theme: every act break exists to advance a specific stage of the argument. Theme is explained by the collision of Character and World: the argument is the specific argument generated when this protagonist's Misbelief meets this world's demands. And so down to Idea. In a draft that almost works, almost all of these explanations hold. One does not. Find that one and you have found the source.
The diagnostic, applied to a near-finished draft, takes one or two evenings. Read the manuscript with the seven questions open beside you. For each layer, ask whether it is explicable by the layer beneath it.
Voice — does the POV serve the structure's informational requirements? If you are writing third limited and the structure requires dramatic irony the limited POV cannot deliver, that is a fault. (It is also rare in almost-works drafts, because voice problems usually surface earlier.)
Structure — do the act breaks correspond to shifts in the thematic argument? Does the midpoint reverse what the theme has been claiming? Does the all-is-lost moment lose something the theme has established as central? If the act breaks are merely dramatic — if they reverse plot rather than argument — that is a fault.
Theme — is the central argument provable from the character-world collision? Read the climax. Ask what it actually demonstrates. Then ask whether what it demonstrates is the argument the story has been building. If the climax resolves the plot without proving the theme, the issue is here. (This is the most common location of the almost-works fault. By a wide margin.)
World — do the world's pressure points target this protagonist's specific psychology? If the world is rich but generic — if the political structure obstructs no one in particular, if the cultural taboos violate no one specifically — the collision will be vague, and the theme generated by it will be vague too. The almost-works softness is downstream of that vagueness.
Characters — does the protagonist carry a specific Misbelief? Are the antagonists engineered to make that Misbelief untenable? If the Misbelief is vague — "she has trust issues" rather than "she believes anyone who has hurt her once cannot be trusted again" — the structure cannot exert specific pressure on it.
Genre — is the genre the right contract for this idea? If the obligatory scenes feel forced, or if the resolution is genre-correct but story-wrong, the fault may be here.
Idea — does the story's premise remain coherent across the manuscript? If the centre has shifted — if the question the story was originally asking has been quietly displaced by a different question that emerged in drafting — the fault is at the bottom. (Rare, but worth the check.)
Work the questions in order. The first one that returns "no, actually" is your fault line. That is where the rebuild begins.
The good news, and it is genuinely good news, is that an almost-works draft does not require a rebuild from scratch. The fault is one layer. Recommit that layer specifically — sharpen the protagonist's Misbelief from generic to specific, revise the world's pressure points to target that specific psychology, name the thematic argument the climax must actually demonstrate — and the recalculation downstream is bounded. Scenes mostly survive. Sentences mostly survive. What changes is what the scenes are now doing — the same prose, doing more specific work, because the architecture beneath it has finally been committed precisely.
The phrase for what tends to happen here is Scribbardiquake — the structural revelation that some things need to move, which sounds alarming but is usually the story becoming what it was always trying to be. The Scribbardiquake of an almost-works draft is small. The architecture mostly held; the fault was localised. The rebuild proves it.
Scribbard's AI Review is the mechanical realisation of this diagnostic. Upload the draft, run a structured review pass, and the system surfaces findings layer by layer — flagged by severity, located by chapter, framed as specific structural questions ("the protagonist's Misbelief is not specific enough to support the climax's argument"; "this subplot does not complicate, illuminate, or contrast the central thematic claim"). The findings are not verdicts. They are diagnostic prompts, each of which can be triaged: marked fixed, dismissed, deferred, or reopened. The pattern in the findings is usually clearer than any single finding — three or four flags clustering around a single layer is the chain breaking, and the rebuild localises there. (You can run the diagnostic-reverse without the tool. The questions are the same. The tool is faster.)
A few practical notes for the rebuild. Do not begin patching prose until the broken layer has been recommitted in writing. The sentences will fix themselves once the architecture beneath them is right; the temptation to start at the prose is the same instinct that produced the polishing-without-progress that probably preceded this revision. Do not run the diagnostic on the manuscript chronologically. Read it with the layers in mind, not the chapters. And do not be discouraged if the broken layer turns out to be Theme. Theme is the most common location, because theme is the layer most often skipped without being noticed; finding it broken is the diagnosis you needed, not a verdict on the work.
The almost-works draft becomes the works draft when one specific layer is recommitted with specificity it did not previously have. The softness around the climax tightens. The ending lands. The middle's weight settles into its proper place. Same novel. Different result. The thing the readers couldn't quite name was a fault line, and the fault line had a name.