17 Drafts to a Published Novel: What I Learned From Iterating The Incubator

The Incubator went through 17 numbered drafts before it was published. The first draft was split across three files totaling about 255 KB. The final published manuscript is 601 KB. If you chart the file sizes across drafts, the line zigzags — it does not climb steadily. The book got shorter before it got longer. It got worse before it got better. Then it got better by getting smaller again.

That trajectory taught me more about editing than any writing workshop.

I'm going to walk through the draft history because I think the specifics matter. Not "draft early and often" — you've heard that. The useful part is seeing where each draft changed the manuscript and why.

Drafts 1-3: Figuring Out What the Book Was

Draft 1 was three separate files: 72 KB, 125 KB, 58 KB. Written over six days in mid-November 2025. It was a first draft in the purest sense — I was discovering the story by writing it. The structure was act-based, inherited from the outline. Characters did things because the plot needed them to.

Draft 2 unified the three files into a single 320 KB manuscript. The act breaks were still there, but the seams showed. This draft got an ePub export, which meant I could read it on a screen like a real book. It immediately felt wrong. The pacing dragged in the middle. Characters introduced in act one disappeared for chapters.

Draft 3 was a structural overhaul at 324 KB — almost the same length, but rearranged. I moved chapters, cut transitions, rebuilt the middle section. The file size barely changed, but the reading experience improved substantially. Same words in a different order is still a different book.

Draft 3 was the first draft that felt like a novel instead of a word count.

Drafts 4-5: The Tightening

This is where it gets interesting. Draft 4 dropped to 260 KB. Draft 5 dropped further to 229 KB. I was cutting. Aggressively. The reasoning was sound — there were scenes that dragged, descriptions that repeated, dialogue that circled the same point three times before landing.

But the cuts went too deep. At 229 KB, the book read fast. Too fast. Scenes that needed room to breathe were compressed to a paragraph. Character motivations disappeared because the setup scenes were cut. The epub generated from Draft 5 felt like a summary of a novel rather than a novel.

The lesson: cutting is necessary. Overcutting is just another form of bad writing. I had traded bloat for thinness. Draft 5 was the worst draft of the seventeen, and it was the shortest.

Drafts 6-9: Rebuilding

Draft 6 bounced back to 337 KB. I wasn't adding new material — I was restoring what should never have been cut. The difference between Draft 5 and Draft 6 was the difference between "remove the redundant paragraph" and "remove the entire scene that establishes why the character cares."

Drafts 7 through 9 were slower, more deliberate work. Draft 7 restructured into acts at 239 KB — another contraction, but this time intentional. The act-based structure forced discipline: each act had a clear purpose, and anything that did not serve that purpose was moved or removed. Draft 8 incorporated an external editor's report at 253 KB. Draft 9 expanded the glossary and character templates at 265 KB.

These four drafts took three weeks. The word count moved in a narrow band — 239 to 337 KB. The changes were structural, not additive. I was shaping the clay, not adding more of it.

Drafts 10-12: The Reader-Critic Loop

Draft 10 jumped to 364 KB. Draft 11 to 420 KB. Draft 12 to 431 KB. This is where the manuscript changed character.

The jump was not me writing new scenes. It was the reader-critic loop. I had built an automated system that read each chapter, flagged consistency issues, identified pacing problems, and suggested where scenes needed expansion. Draft 10 incorporated the first round of reader-critic feedback: expanded chapters 10 through 17, continuity fixes across character arcs.

Draft 11 integrated a second, deeper pass. Fourteen chapters were revised based on specific feedback: "this character's motivation is unclear in chapter 12," "the transition between chapters 14 and 15 breaks the emotional arc," "this dialogue establishes a relationship that contradicts chapter 7."

The reader-critic loop did what no human beta reader could do: it read every chapter against every other chapter simultaneously and found the contradictions.

The word count climbed because the feedback identified real gaps. Scenes that should have existed. Transitions that were missing. Character moments that had been implied but never shown. This was not padding — it was the connective tissue the book needed.

Drafts 13-16: Per-Chapter Files

At Draft 13, I switched from a single manuscript file to 27 individual chapter files (Ch00 through Ch26). The total was about 477 KB. Drafts 14, 15, and 16 refined individual chapters — 493 KB, 496 KB, 498 KB. The growth was small and targeted.

This format change was the most productive structural decision of the entire process. When the manuscript was one file, every edit required scrolling to find the section, making the change, and then reading surrounding chapters to check for consistency. With per-chapter files, I could open chapter 12, make a targeted revision, and run the consistency checker against just that chapter.

The per-chapter approach also made the iteration visible. I could see which chapters had been touched in each draft and which had not. Draft 15 only modified chapters 7, 10, 12, 18, and 23. Five chapters out of twenty-seven. That is not a full draft pass — it is surgery.

Working at the chapter level turns "another draft" from a daunting rewrite into a series of focused edits.

Draft 17: The Final Cut

Here is the counterintuitive part. Draft 17 — the final draft before production — was smaller than Draft 16. It dropped from 498 KB to 444 KB. The final edit was a cut.

By Draft 16, I had been staring at the manuscript for five weeks. I knew which scenes dragged. I knew which descriptions repeated information the reader already had. I knew which dialogue was the character saying what I wanted them to say instead of what they would actually say. But knowing something is wrong and being willing to cut it are different things.

Draft 17 was the draft where I was finally willing. I cut 54 KB — roughly 13,000 words. Scenes that I loved but that did not serve the story. Beautiful descriptions that slowed the pacing. A subplot that went nowhere.

The published version at 601 KB is larger because of formatting, front matter, and back matter. The manuscript itself, stripped of production additions, is essentially Draft 17. The cuts held.

The last draft should remove things. If your final edit is adding material, you are not done. You are still building. The final edit is the one where you trust what is there and remove what is not needed.

What I Would Do Differently

Looking at the progression, the most wasteful drafts were 4 and 5. I cut too aggressively, then spent drafts 6 through 9 rebuilding what I had removed. If I had been more surgical in the initial cuts — removing specific paragraphs rather than entire scenes — I could have skipped four drafts.

The most valuable drafts were 10 through 12, the reader-critic loop. That is where the manuscript went from "a complete novel" to "a good novel." The feedback was specific, consistent, and grounded in cross-chapter analysis that human readers do not naturally do.

The per-chapter switch at Draft 13 should have happened at Draft 7. Working in a single large file for six drafts was a structural tax on every edit.

The Principle

The general principle: iteration is not linear. The word count chart for The Incubator goes up, down, up, up, up, down, up, up, sideways, down. Each movement was a response to a specific problem. The drafts that increased word count were filling gaps. The drafts that decreased word count were removing what did not belong. Both directions were productive.

The instinct to "just keep writing forward" is wrong. So is the instinct to "just keep cutting." The productive direction depends on what the manuscript needs at that moment, and the only way to know is to read it critically, find the specific problem, and fix that specific problem. Seventeen times.