The direct answer: every scene needs one clear job
Enter each scene with one purpose: plot, character, setup, or world. The validator scores every entry, measures purpose coverage across all four jobs, flags consecutive filler, and labels each scene Keep, Combine, or Cut. You get an overall keep rate so you know which scenes to cut first.
That is the whole method on this page. You list the scenes in order, you assign each one the single strongest job it actually performs, and the tool tells you which scenes earn their place and which ones only occupy space. Scenes that move the story or deepen a person score highest and stay. Scenes that plant information or establish the setting score in the middle and should be folded into a stronger neighbor. Scenes with no real job score zero and should be deleted, with any salvageable line carried into a scene that survives.
Most drafts do not suffer from one terrible chapter. They suffer from five or six quiet scenes that each feel harmless alone and together drain the middle of the book. This check exists to surface those scenes by name, in order, with a decision attached to each. Run it on any sequence that feels slow: an opening, a sagging middle, or the run-up to the climax. Fix what it flags, then read the pages aloud to confirm the rhythm returned.
What the four purposes actually mean
A purpose is the job a scene performs for the reader, not the activity the characters perform inside it. Two characters eating dinner can carry any of the four purposes depending on what changes by the final line. The validator asks you to name one purpose per scene so the scoring stays honest. When you feel tempted to assign two purposes, pick the stronger one and treat the weaker one as a bonus, because scenes that claim two jobs often perform neither.
Plot means the story situation changes in a way later scenes must react to. An arrest, a betrayal, a discovered letter, a missed train, a confession that forces a choice: these move plot. The test is simple. If you removed the scene, would the events of the next chapter still make sense unchanged? When the answer is no, the scene carries plot. Plot scenes score 3 and receive a Keep decision because they are load-bearing. A manuscript can survive weak sentences inside a plot scene, but it cannot survive the removal of the plot turn itself.
Character means our understanding of a person changes in a lasting way. A quarrel that reveals what someone fears, a kindness that contradicts a reputation, a failure that exposes a limit: these deepen character. The test is whether the reader would describe the person differently after the scene than before it. Notice that feelings alone do not satisfy this test. A scene where someone feels sad but nothing is revealed and nothing is decided is not a character scene. It is a mood, and moods without revelation score as filler. True character scenes score 3 and receive a Keep decision.
Setup means the scene plants information that a later scene pays off. A letter is mentioned, a rule is stated, a weapon is shown, a debt is recorded, an appointment is made. Setup scenes often feel quiet because their payoff arrives chapters later. The test is whether you can name the specific later scene that uses what was planted. When you can name it, the scene carries setup and scores 2 with a Combine decision. When you cannot name it, the planted detail is decoration and the scene is closer to filler than you hoped. Be strict here, because hopeful setup is the most common disguise filler wears.
World means the scene establishes the rules, costs, or pressures of the setting in a way that constrains future choices. Harbor laws that limit when ships may leave, a guild tithe that punishes late payment, a winter road that closes at first snow: these are world. The test is whether the rule or pressure changes what characters can attempt next. Description alone does not qualify. A beautiful market street that constrains nothing is backdrop, and backdrop without pressure scores as filler. Genuine world scenes score 2 with a Combine decision, because setting pressure usually lands hardest when delivered inside action that already matters.
None means the scene performs none of the four jobs at a level the story can feel. Travel montages that change nothing, meals that reveal nothing, rest stops that plant nothing, banter that constrains nothing: these score 0 and receive a Cut decision. Labeling a scene none is not an insult to your writing. Many cut scenes contain fine sentences. The label only means the story would lose nothing if the scene vanished, and fine sentences can be relocated into scenes that survive.
How the scoring math works, step by step
The tool applies fixed weights to your purpose labels. Plot scores 3. Character scores 3. Setup scores 2. World scores 2. None scores 0. There is no judgment of your prose quality inside these numbers. The numbers only encode a craft convention: scenes that change the situation or the person carry a sequence alone, while scenes that plant information or establish rules usually need a stronger neighbor to carry them.
Each scene also receives a decision from its score. Score 3 becomes Keep. Score 2 becomes Combine. Score 0 becomes Cut. The decision vocabulary is deliberately small so you can act on it in one revision pass. Keep means leave the scene standing and revise at sentence level. Combine means fold the scene into the nearest Keep scene and deliver its information inside that action. Cut means delete the scene and rescue at most one or two lines into a surviving scene.
Coverage counts how many of the four real purposes appear at least once across the entered sequence. Plot, character, setup, and world each count once no matter how many scenes carry them. A sequence with plot, character, and setup present has coverage 3 of 4, and the missing purpose is named. Coverage matters because long stretches that never touch one purpose start to feel thin. A middle that never advances plot feels static. A middle that never deepens anyone feels mechanical. A middle that never plants anything feels episodic. A middle that never lets the setting push back feels weightless.
Filler count is the number of scenes labeled none. Keep rate is the share of scenes that carry any real purpose, computed as keep count divided by total scenes, multiplied by one hundred and rounded to a whole number. Consecutive filler pairs count the number of adjacent positions where two none scenes sit side by side. That last measure exists because placement matters as much as quantity: one rest stop between strong scenes reads as breathing room, while two rest stops in a row read as stalling.
Worked numbers from the default five-scene list
Walk through the default input to see every calculation. The list is Dock arrival labeled plot, Market quarrel labeled character, Inn rest labeled none, Arrest labeled plot, and Letter labeled setup. That is five scenes in entered order.
Scores fall out directly from the weights. Dock arrival earns 3. Market quarrel earns 3. Inn rest earns 0. Arrest earns 3. Letter earns 2. The total is 11 points from a possible 15, but the tool does not grade you on points. Points only determine per-scene decisions: Keep, Keep, Cut, Keep, Combine.
Keep count is 4, because four scenes carry a real purpose and one does not. Filler count is 1. Keep rate is 4 divided by 5, which equals 80 percent. That single percentage is the fastest summary of the sequence: four fifths of the scenes earn their place.
Coverage is 3 of 4. Plot appears through Dock arrival and Arrest. Character appears through Market quarrel. Setup appears through Letter. World never appears, so the tool names world as the missing purpose. For a fantasy sequence this gap deserves attention, because the genre leans on setting pressure, and a five-scene run with no world purpose may feel unanchored even though the keep rate looks healthy.
Consecutive filler pairs equal 0, because the single none scene sits alone between stronger scenes. That placement is the kindest possible case for filler: Inn rest reads as a breath between the quarrel and the arrest rather than a stall. The recommendation therefore starts with cutting or absorbing Inn rest, continues with combining Letter into Arrest or into Dock arrival so the planted information travels inside action, and ends with giving one surviving scene an explicit world job so coverage reaches 4 of 4.
The overall verdict for this input is a warning, not a failure. Eighty percent keep with one isolated filler and one coverage gap is a normal first-draft shape. It needs one deletion, one fold, and one added beat, which is an evening of revision rather than a rebuild.
A second worked calculation with heavy filler
Consider a struggling middle entered as Rest labeled none, Meal labeled none, Nap labeled none, and Arrest labeled plot. Four scenes, three of them filler.
Scores are 0, 0, 0, and 3. Keep count is 1. Filler count is 3. Keep rate is 1 divided by 4, which equals 25 percent. Coverage is 1 of 4, because only plot appears. Consecutive filler pairs equal 2: Rest plus Meal form one pair, and Meal plus Nap form a second pair, since the counter advances once per adjacent position.
The verdict for this input is a failure, and the number explains why readers report putting the book down in this stretch. Three consecutive rest scenes deliver no change, no revelation, no planted payoff, and no setting pressure before the arrest finally moves. The fix is structural, not stylistic. Delete at least two of the three filler scenes outright, move any beloved line into Arrest, and give Arrest or its neighbor a character beat and a planted detail so the sequence after the cut carries three purposes instead of one. No amount of sentence polishing on Nap will substitute for that deletion.
Compare the two calculations and the logic of the tool becomes visible. The first list needs tightening. The second list needs surgery. Both diagnoses come from the same four counts, which is why entering honest labels matters more than entering many scenes.
How to enter your scenes so the scores mean something
Type each scene as a name, a colon, and a purpose word, with commas between entries. A correct entry looks like Dock arrival:plot in structure, using your own scene names and one of the five allowed purpose words. The five allowed words are plot, character, setup, world, and none, all in lowercase. Anything else in the purpose position is treated as none, which is intentional: a purpose you cannot name with the shared vocabulary is not yet a purpose.
Keep scene names short and recognizable. Three to five words is enough. You need to identify the scene in the recommendation list later, so Harbor negotiation or Letter from the sister beats a vague label like Scene 7. Order matters, so enter scenes in manuscript order rather than importance order. Consecutive filler detection only works when the list mirrors the reading experience.
Assign the single strongest job, not every job the scene touches. If a scene advances the plot and also reveals character, decide which change the reader would notice first and label accordingly. When the choice feels genuinely close, prefer plot or character over setup or world, because the stronger claim is the one the scene must defend to earn a Keep. If no claim survives scrutiny, label the scene none. Honest none labels are the most valuable input you can give the tool, because every hidden filler you relabel as setup to spare its feelings corrupts the coverage count and hides the cut you need.
Limit a single run to the sequence you are actually revising. Five to twelve scenes is the useful range. Below five, coverage math is noisy: a short list can miss a purpose without any real problem. Above twelve, the recommendations blur because distant scenes cannot sensibly combine with each other. For a full manuscript, run the check chapter by chapter or movement by movement, and collect the Cut decisions into one deletion list at the end.
Watch for the most common entry errors. Leaving off the colon turns the whole entry into a name with purpose none, which punishes a good scene for a punctuation slip. Using inventive purpose words like tension or backstory also resolves to none, so stick to the five allowed words. Trailing commas and extra spaces are harmless and ignored. An empty input produces an informational result that asks for scenes rather than a fake score, so a blank run never flatters you.
Illustrative hypothetical example: a mystery opening that keeps too much setup apart
This first illustrative hypothetical example follows an invented mystery writer revising her opening five. Her list reads: Body found labeled plot, Detective golfs labeled none, Neighbor interview labeled character, History of the manor labeled world, and Missing will mentioned labeled setup. Five scenes, entered in order.
Scores come out as 3, 0, 3, 2, and 2. Keep count is 4, filler count is 1, and keep rate is 80 percent. Coverage is 4 of 4, because every purpose appears at least once: plot in the discovery, character in the interview, world in the manor history, setup in the missing will. Consecutive filler pairs equal 0, since the golf scene stands alone.
The tool returns a warning rather than a pass, and the reasoning shows how decisions differ from scores. Coverage is complete and the rate is healthy, yet one scene still scores zero. Detective golfs is a pleasant character moment in the author's mind, but under the strict test it reveals nothing lasting and changes nothing downstream. The recommendation names it as the cut: delete the golf outing, and if one line of voice is worth saving, move that line into the neighbor interview where a Keep scene can carry it.
The two Combine scenes need folding, not deletion. History of the manor delivers genuine setting pressure, since inheritance rules constrain who could benefit from the death, but as a standalone lecture it slows the opening. The fix is to deliver the manor rules inside the neighbor interview: let the neighbor cite the entail clause while defending her own claim, so world travels inside character. Missing will mentioned plants the key payoff, so it should move into the body-found scene or the interview rather than standing as its own errand. After one cut and two folds, five scenes become three, the opening carries all four purposes in three chapters, and the investigator reaches the first real choice fifty pages sooner.
Notice what this hypothetical revision did not do. It did not rewrite a single sentence of the surviving plot scenes. Structure first, sentences second: that order is the entire discipline this tool teaches.
Illustrative hypothetical example: a fantasy middle drowning in travel
This second illustrative hypothetical example follows an invented fantasy writer stuck in his middle eight. His list reads: Ford crossing labeled world, Campfire meal labeled none, Ridge walk labeled none, Toll dispute labeled plot, Night watch labeled none, Oath remembered labeled character, Map copied labeled setup, and Gate denied labeled plot. Eight scenes, entered in order.
Scores are 2, 0, 0, 3, 0, 3, 2, and 3. Keep count is 5, filler count is 3, and keep rate is 5 divided by 8, which rounds to 63 percent. Coverage is 4 of 4, since world, plot, character, and setup each appear. Consecutive filler pairs equal 1: Campfire meal plus Ridge walk sit adjacent, while Night watch stands alone between Toll dispute and Oath remembered.
The verdict is a failure, because a 63 percent keep rate sits near the threshold where readers describe a middle as slogging even when individual chapters read well. The recommendations attack placement first. The adjacent pair must be broken: delete Ridge walk entirely and absorb Campfire meal into Ford crossing, so the meal conversation happens while the party is still cold from the river and the world pressure of the current gives the dialogue something to push against. Night watch, the isolated filler, is the next cut: move the one atmospheric image worth keeping into Gate denied and delete the rest.
Then the Combine scenes get folded. Map copied plants the route detail that Gate denied pays off, so the copying should happen inside Toll dispute, where a Keep scene with conflict can carry exposition without slowing. Ford crossing, the world scene, should merge into Toll dispute as well or into Gate denied, so the river rules and the gate rules read as one continuous system of control rather than two separate lessons. Oath remembered already keeps as character, but it can also carry the emotional cost of the deleted Night watch by letting the remembered oath interrupt sleep in a single paragraph inside a surviving scene.
After revision, eight scenes become four: a combined crossing and meal, a toll dispute that also copies the map, an oath scene that absorbs the watch, and a gate denial that pays off the map. Keep rate rises to 100 percent across the survivors, coverage stays complete, and consecutive filler disappears. Nothing in this hypothetical required inventing new plot. It required deleting three scenes and relocating their salvageable lines, which is the characteristic move this validator keeps prescribing.
Keep, Combine, Cut: what each decision asks you to do
Keep asks for sentence-level revision only. The scene stands where it is. Tighten entrances and exits, sharpen the turn, cut throat-clearing lines at the start, and land the final line on the change. Do not restructure a Keep scene to fix a neighboring Cut scene. Let the strong scenes hold still while you move weaker material into or out of them.
Combine asks for a fold. Identify the nearest Keep scene that can host the weaker scene's information, move the essential beat across, and delete the husk. The most common folds pair setup into plot and world into character. A letter is better read during an arrest than during an errand. Harbor law is better cited during a quarrel than during a tour. When choosing the host, prefer the nearest Keep scene in reading order so the information arrives at roughly the same point in the story. Moving setup three chapters early can spoil; moving it three chapters late can confuse.
Cut asks for deletion with salvage. Delete the scene from the manuscript, then rescue at most one or two lines: an image, a joke, a detail of voice. Place each rescued line inside a surviving scene where it does not change the host's purpose. If you find yourself rescuing whole paragraphs, the scene may have been mislabeled and deserves a second look. Relabel it honestly and rerun before you restore it, because restored scenes have a habit of dragging their filler back in with them.
A practical order for applying decisions: cut first, combine second, rebalance coverage third. Cutting first shortens the list and often resolves consecutive filler pairs automatically. Combining second folds the middle-weight scenes into the survivors. Rebalancing third adds the missing purpose to one surviving scene as a second beat rather than building a new scene from scratch. Check the keep rate after each stage so you can see the number move and stop revising when the sequence passes.
Why consecutive filler hurts more than scattered filler
Readers forgive an occasional quiet scene. They rarely forgive two in a row. A single rest stop between strong scenes reads as contrast: the quarrel lands harder because the meal before it was calm. Two rest stops in a row read as intention: the story appears to have stopped caring about its own momentum, so the reader stops caring too.
The tool counts consecutive pairs separately from total filler for this reason. A list with two scattered none scenes and a list with two adjacent none scenes can share the same keep rate while producing very different reading experiences. The scattered version needs two small deletions. The adjacent version needs emergency surgery on that exact neighborhood, because the stall compounds: the second filler inherits the impatience created by the first.
When the recommendations name a consecutive pair, treat the pair as one problem. Cut the weaker of the two entirely, then decide whether the stronger deserves a Combine by gaining a purpose or a Cut as well. Never solve a consecutive pair by adding a new plot scene between the two fillers. Inserting action between two stalls preserves both stalls and lengthens the slow stretch. Remove before you add, always.
Openings tolerate consecutive filler worst of all. Readers have no invested momentum to spend in chapter one, so two quiet scenes before the first turn can end the book before it starts. Middles tolerate slightly more, provided the filler sits between major turns rather than after a cliffhanger. Endings tolerate almost none: once the final turn begins, any scene without plot or character purpose reads as delay rather than breath.
Edge cases and failure modes to handle honestly
Short lists produce noisy verdicts. A two-scene prologue that carries plot and character will show coverage 2 of 4 and a warning about missing setup and world. That warning is arithmetic, not a command. Coverage is a sequence-level measure, and two scenes cannot reasonably carry four purposes. Restrict coverage judgments to runs of five or more scenes, and use shorter runs only for per-scene decisions.
Finale sequences skew toward plot and character by design. A four-scene climax with three plot scenes and one character scene may show setup and world missing with a 100 percent keep rate. Do not insert a world lecture into the climax to satisfy the counter. Missing purposes in the finale usually indicate that setup and world did their work earlier, which is correct. The tool names the gap; your judgment decides whether the gap belongs to this sequence or to an earlier one.
Montages and travel sequences often get mislabeled as world when they are actually none. Crossing three towns is not world purpose unless a rule or pressure constrains the crossing. A toll, a closed pass, a curfew, a debt collected at the border: these convert travel into world. Without constraint, travel is movement without meaning, and labeling it world only hides the cut. When in doubt, ask what the travelers cannot do because of this place. If the answer is nothing, label the scene none.
Dialogue-heavy scenes get mislabeled as character when they are actually setup wearing conversation. Two people exchanging information they both already know is exposition, not revelation. The character test requires a lasting change in understanding, not merely strong feelings during the talk. If the reader's description of either participant is identical before and after, the scene is setup at best. Score it 2, combine it into action, and stop defending it as a 3.
Prologues and epilogues break the keep-rate heuristic when judged alone. A prologue that plants a single setup detail across four pages will score Combine or Cut in isolation while performing its function for the whole book. Judge frame scenes against the book, not against the run: if the prologue's planted detail pays off in the climax, keep it regardless of the local score. The validator is a sequence tool, and frames belong to the full structure.
False precision is the main failure mode of any scoring aid. Scores of 3, 2, and 0 are coarse by design, and keep rate is a rounded whole number. Do not compare 83 percent against 86 percent across runs as though three points of difference carried meaning. Read the bands instead: 100 percent with full coverage passes, 80 percent and above needs tightening, below 60 percent needs surgery, and anything with consecutive filler needs placement repair regardless of band.
Gaming the labels is the second failure mode. Relabeling a rest scene as setup without naming the payoff scene raises the keep rate and silences the warning while leaving the slow pages untouched. The number improves and the book does not. Defend every setup label by writing down the payoff scene it serves. If you cannot name one, accept the none and take the cut.
Genre-specific guidance for the four supported genres
Genre changes which missing purpose should worry you most. The tool watches one focus purpose per genre, but the reasoning below explains how to weigh every gap in context. These are craft conventions observed across many books, not measurements of any market, and they describe reader expectations, not rules.
Fantasy runs on setting pressure. Readers arrive expecting rules with costs: magic with prices, guilds with dues, roads with seasons, courts with protocols. A fantasy sequence missing world purpose feels unanchored even when plot moves briskly, because actions seem to happen in a void rather than inside a system. When the tool flags missing world in fantasy, give one surviving scene a constraint beat. Let the harbor master cite the tide law during the quarrel. Let the toll keeper name the guild cut during the dispute. One or two constraint beats per movement usually restore the sense that the world pushes back. Fantasy also tolerates setup better than other genres, because readers accept planted rules and names, but only when each plant pays off visibly later. Audit every fantasy setup label against its payoff before defending it.
Mystery runs on planted information. Readers expect clues, timelines, alibis, and physical details that later assemble into the solution. A mystery sequence missing setup purpose starves the ending, because the reveal will feel unearned no matter how dramatic the arrest scene is. When the tool flags missing setup in mystery, plant one verifiable fact inside a Keep scene: a timetable glimpsed during the interview, a muddy boot noticed during the quarrel, a letter referenced during the arrest. Mystery also demands strict treatment of none scenes. Detective rest scenes that reveal nothing and plant nothing read as stalling between clues, and consecutive filler in the investigation middle is the classic cause of readers skipping ahead. Cut rest scenes aggressively and let the investigator think inside active scenes instead.
Romance runs on character turning points. Readers expect the relationship to change what each person wants, fears, or will admit. A romance sequence missing character purpose feels like motion without feeling: dates, trips, and quarrels occur, but nobody's understanding shifts. When the tool flags missing character in romance, give one surviving scene a revelation beat. Let the quarrel expose the fear behind the demand. Let the quiet errand force a choice between pride and honesty. Romance also uses setup differently: plant the future obstacle rather than a clue. A mentioned ex-partner, a stated boundary, a visible habit. Each plant should convert into a relationship test later, or it is decoration.
Thriller runs on plot velocity. Readers expect the situation to worsen, narrow, or reverse at short intervals. A thriller sequence missing plot purpose feels like preparation without danger: meetings, travel, and research accumulate while nothing gets worse. When the tool flags missing plot in thriller, convert one surviving scene's ending into a turn. Let the interview produce an arrest warrant. Let the request for help produce a refusal with a deadline. Thriller tolerates world and setup only in small doses inside action: a rule cited during pursuit, a planted detail glimpsed during escape. Standalone setup lectures and setting tours stall thrillers faster than any other genre, so combine them without mercy.
Cross-genre books should satisfy the focus of their shelving genre first. A romantic mystery must plant clues even though its heart is character. A fantasy thriller must keep plot turning even though its joy is world. When two genre expectations conflict, let the primary genre's focus win for coverage and treat the secondary focus as a bonus beat inside Keep scenes.
A revision workflow that uses the results in one sitting
Start with a printed or duplicated list of the tool's table, in entered order, with decisions visible. Mark every Cut scene with a single line through its name and every Combine scene with an arrow toward its host Keep scene. Do not touch the manuscript yet. Planning the moves on the list prevents the common error of deleting inside the draft and losing track of what the survivors must absorb.
Delete the Cut scenes from the draft first, rescuing at most two lines per scene into their hosts. Read the seams where deletions occurred and smooth only the transitions: a sentence of orientation at most. Resist the urge to polish surrounding chapters at this stage, because the sequence is still mid-surgery and polish will be wasted on sentences that may move again.
Fold the Combine scenes next. Move each setup plant or world constraint into its host, choosing the smallest possible insertion that preserves the information. A letter can be read in two sentences inside an arrest. A harbor law can be cited in one line inside a quarrel. After each fold, delete the husk scene completely and confirm the host still reads as itself rather than as two scenes stitched together. When a host starts to feel crowded, the fold target was wrong: choose the other neighboring Keep scene instead.
Rebalance coverage last. Look at the missing purpose named in the metrics and add it as a beat inside one surviving scene. A missing world beat becomes a cited rule with a cost. A missing setup beat becomes a mentioned detail with a named payoff. A missing character beat becomes a revealed motive inside an existing exchange. A missing plot beat becomes a worsened situation in the final line. One beat is enough; coverage counts presence, not quantity.
Rerun the validator on the revised list with honest labels. Confirm the keep rate rose, the consecutive pairs reached zero, and coverage shows 4 of 4 or a defensible gap you can explain. Then read the surviving pages aloud from the first scene through the last. Listening catches transition problems that numbers cannot: repeated orientations, doubled goodbyes, time jumps that need one sentence of bridge. Fix those, and the sequence is done.
Questions writers ask, answered in depth
How strict should the character test be? Strict enough that a scene passes only when you can state the changed understanding in one sentence. She learned he fears debt more than danger. He admitted the marriage is over. She chose the crew over her pride. When you can write that sentence, the scene earns its 3. When the sentence comes out as she felt sad but stayed the same, the scene is atmosphere, and atmosphere belongs inside a Keep scene as a paragraph, not standing alone as a chapter.
Can a scene carry setup for a payoff much later? Yes, provided the plant is specific and retrievable. A named letter, a stated rule, a shown object, a recorded debt: these survive in reader memory because they are concrete. Vague foreboding does not survive. A sense that something is wrong plants nothing, because no later scene can point back to it. Score concrete plants as setup and vague unease as none, and your coverage count will start telling the truth.
Is one filler scene ever acceptable? One isolated none scene inside a strong sequence is the most forgivable shape filler takes. A brief rest after a major turn can regulate pacing, and some readers welcome a breath. The conditions are narrow: exactly one, isolated on both sides by Keep scenes, short, and never adjacent to another none. Two filler scenes anywhere near each other, or one long filler that consumes a full chapter, forfeits this tolerance. When the tool shows a single isolated Cut with no consecutive pairs and high keep rate, you may keep it deliberately, but shorten it by half and give it one sensory detail worth the pause.
Should setup scenes ever stand alone? Rarely, and only when the plant requires ceremony the story will reuse. A will reading, a rule proclamation, a map presentation: these can justify their own scene when later chapters quote them back. Even then, attach a character reaction strong enough that the scene could defend a second label. Pure information delivery without reaction is the shape most likely to be skipped, so treat standalone setup as guilty until proven ceremonial.
What if cutting filler shortens the book too much? Length built from filler was never length; it was delay. A shorter manuscript where every scene keeps will find readers more easily than a longer one where a quarter of the scenes stall. When a contracted sequence leaves a gap in the arc, add plot or character work to surviving scenes rather than restoring a Cut. Extend the arrest into a choice. Extend the quarrel into a revelation. Growth inside Keep scenes adds pages that earn their place, while restored filler adds pages that repeat the original complaint.
How do I handle disagreement with a Cut decision? Argue with the label, not the score. The score follows the label mechanically, so a disputed Cut is always a disputed none. State the purpose you believe the scene carries and apply that purpose's test: name the changed situation, the changed understanding, the payoff scene, or the constraint. When the evidence holds, relabel and rerun. When the evidence dissolves under questioning, the tool was right and the scene goes. This habit of defending labels out loud is the fastest way to develop editorial judgment that persists after you close the page.
Where to draft the revised sequence
When your decisions are made, open Co-Writer at /write and draft the revised scene list there, starting with the Keep scenes in their surviving order. Work through one flagged scene at a time inside the same draft so each new purpose is tested on the page rather than only in the list. Keep the validator results beside you while you revise so every Combine choice turns into a concrete added beat instead of a vague intention.
Quick reference before you run the tool
List five to twelve scenes in manuscript order using name-colon-purpose entries separated by commas. Label each scene with its single strongest job from plot, character, setup, world, or none, and treat any invented sixth word as none. Expect Keep for plot and character, Combine for setup and world, and Cut for none. Read keep rate as the headline number, coverage as the completeness check, and consecutive pairs as the pacing alarm. Cut first, combine second, rebalance coverage third, then rerun and read aloud. Sequences maintained this way stay lean across drafts, because every future scene must answer the same question before it is allowed to stay: what is your job, and which later scene would break without you.