Your file is not finished when the writing is finished
A manuscript that looks right in Word is not necessarily right for the machine that will turn it into a book. Word is a screen writer; KDP Print is a paper press and EPUB is a reflow stream. What survives from one to the other is not what you see on screen but what you named. Unnamed formatting looks bold and centered on your monitor and does not have a name the converter can carry. That is why a file that looks clean on your monitor can still come back from a formatter with a markup bill.
The converter does not see bold. It sees a style named Strong or Heading 1 that happens to be bold. When you apply bold directly to text without a style, the text is bold in the visual layer but carries no name in the style layer. When the template maps styles to the printer's fonts, the direct bold has no name to match and falls through. It may stay bold, it may not, and it will not stay bold consistently across the chapters that were formatted at different moments with different manual tweaks. The fix is boring: give every piece of formatting a name first, then let the name carry the formatting.
This is also why a clean manuscript is faster to format than a pretty one. A pretty manuscript with manual tabs, double spaces, empty paragraphs, and mixed quotes takes a formatter longer to clean than a plain manuscript that uses Normal and Heading 1 correctly and nothing else. The pretty file looks closer to a book on screen and is further from a book in the machine. The plain file looks further from a book on screen and is closer to a book in the machine.
The eight issues that block a clean conversion
Every book that comes back with a conversion invoice shares one or more of eight issues. They are not exotic. They are habits from drafting.
1. Direct formatting instead of styles. Word has two layers: the style layer and the direct layer. The style layer is a named rule — Normal is 11 on 14 Garamond with first-line 0.30 — and the direct layer is a manual override — select this paragraph and make it 11 on 14 Garamond with first-line 0.30. Both look identical on screen. Only the first has a name. The checklist flags direct formatting because it is the only formatting that cannot be mapped. The fix is to create the styles first — Normal, First Line Indent, Block, Heading 1, Heading 2 — apply them where direct formatting was used, and then clear direct formatting with the styles pane command so nothing remains in the direct layer.
2. Manual tabs or spaces for first-line indent. Drafters press Tab or a few spaces at the start of each paragraph to push the first line in. The tab is a fixed character with a fixed width; the spaces are literal space characters that survive reflow. On a new trim the fixed width is wrong, and on EPUB reflow the spaces collapse or do not. The formatter must delete every leading tab and space and replace them with a paragraph style that carries 'first-line' as a rule. The checklist flags manual tabs because they are the single most common source of uneven indents in a proof. Do not fix this by replacing Tab with a first-line style on top of the Tab. Remove the Tab, then apply the style.
3. Double spaces after periods. Two spaces where one is expected create rivers in justified interior type — pale vertical gaps where word spaces align. Word will show them, InDesign will show them more, and the justified PDF will make them louder. They are also easy to miss by eye. The checklist flags double spaces because the find operation is trivial — find two spaces, replace with one — and because the result should be zero. Do this after the tab pass; tabs followed by spaces can reintroduce doubles.
4. Double paragraph returns — empty paragraphs between blocks. Pressing Enter twice to get space between blocks creates an empty paragraph — a paragraph with no text. In print that empty paragraph has a height that participates in widow and orphan calculation, so it can push a heading onto a new page or strand a line. In EPUB the empty paragraph is theme-dependent; some readers hide it, some show it as a gap, and some do not. The checklist flags double returns because the formatter must delete every empty paragraph and replace it with 'space-after' in the style. The visual result is the same — space between blocks — but the machine result is correct: one paragraph with space after, not two paragraphs with one empty.
5. Mixed straight and curly (smart) quotes. Word can produce both. Smart quotes are '“ ”' and '‘ ’'; straight quotes are '" "'. When drafting moves between machines, keyboards, or pasted research, both appear. The mixed file has some dialogue opening with '“' and some with '"', and some apostrophes as '’' and some as '''. The PDF will show the mismatch visibly — straight quotes in a curly interior look wrong — and the EPUB may encode the straight set differently. The checklist flags mixed quotes because the normalize pass is a single operation: enable smart quotes and find/replace straight quotes after confirming pair direction. Do this after the structure pass so you do not reintroduce straights inside the deletions.
6. Hyphen where em dash or en dash is needed. The hyphen-minus '-' is not the em dash '—' or the en dash '–'. Writers typing quickly use '-' for interruption where '—' is correct and for ranges where '–' is correct. The hyphen is visibly too short for interruption and too short for a range em, and search treats hyphen differently. A find for ' - ' where ' — ' is intended catches most interruptions. The checklist flags hyphens because they are the cheapest typographic tell that a file has not had a style pass, and they survive into both PDF and EPUB unchanged.
7. Manual page numbers or typed header/footer. Manuscripts with page numbers typed into the footer, or chapter numbers typed as "Page 3 — Chapter 2", carry a snapshot of a pagination that no longer exists once trim or leading changes. The formatter must delete every manual number and place a real numbered section. The checklist flags this because it is the only issue that becomes wrong the moment any other issue is fixed — change the trim and every manual number is wrong.
8. Low-res or inline-heavy images. Images pasted from the web at 72 or 96 dpi pixelate at the 300 dpi required for print PDF; images pasted inline with text wrapped as "in line with text" create layout problems when the interior reflows around them for EPUB. The checklist flags images because the fix requires re-sourcing at 300 dpi and re-anchoring with text wrap off, which is a different kind of work — asset work — that should not be discovered at export.
Why order matters — styles → structure → characters → pagination
The checklist recommends a fix order because each step can reintroduce the issue the next step fixes if done out of order.
Styles first: map every piece of direct formatting to a named style and clear direct formatting. This is the foundation; every later step depends on text having a style to hold the replacement.
Structure next: remove manual tabs and the spaces at paragraph starts and delete empty paragraphs. Set 'first-line' and 'space-after' in the style after the removal. If you do characters before structure, the double-space find will replace inside the space characters that belong to manual indents you are about to delete, which wastes a pass.
Characters next: find double space → single, normalize smart quotes, and fix em dashes. Do this after structure so the characters you normalize belong to the final paragraph structure, not to throws that will be deleted.
Pagination last: remove manual page numbers and headers, then let the engine paginate. If you paginate before the style pass, the page numbers you place will be wrong after the style pass changes line breaks.
This order is also why the tool estimates time as 'wordCount × issues'. A 75,000-word file with three issues at 6 minutes per 10k is 45 minutes; a 150,000-word file with four issues at 6 minutes per 10k is 90 minutes. The per-10k grows with issues because each issue is a distinct pass, not a single find that fixes everything.
Format changes what hides
Word, Google Docs, and Scrivener each hide different issues. Word shows double spaces and smart quotes well; Google Docs collapses some double spaces in the editor and handles smart quotes differently on export; Scrivener hides the distinction between direct and style formatting until it compiles. Exporting Google Docs to '.docx' before the cleanup pass makes hidden issues visible. Compiling Scrivener to '.docx' before the cleanup pass has the same effect. The tool takes the source format as input so the estimate includes the export reveal — Google Docs to Word often adds 10 minutes of re-finding because the exported file reintroduces a mixed state.
How the formatter sees the file
A formatter receives a file and runs three invisible checks before quoting: style audit (are there styles to map?), structure audit (are there tabs, doubles, empties?), and character audit (are quotes, dashes, and spaces consistent?). A file with zero issues on those three passes goes directly to template mapping. A file with three issues goes through three manual find/replace rounds before mapping. The invoice reflects the rounds, not the page count. The checklist is the same three audits with names and a fix sentence per issue, so the author can run them in Word before the formatter runs them in InDesign or the EPUB engine.
Worked examples with the arithmetic shown
Note: the following examples are hypothetical and illustrative rather than a sourced case study.
A 75,000-word Word file with tabs, mixed quotes, and direct formatting. Word count 75,000 at '.docx' means the tool takes 'ceil(75000/10000)=8' blocks. Three issues ticked at 6 minutes per 10k is 18 minutes per block, or '8 × 18 = 45' quoted as '45 min' by the ceiling rule that rounds up to the nearest 5 and clamps to 15 min minimum. Verdict is 'fail' — "3 issues — clean the file before formatting" — because three checked issues exceed the two-issue warning band. The checklist shows eight items with five done (double spaces, double returns, em dashes, page numbers, images are not ticked and therefore done) and three not done (direct formatting, manual tabs, mixed quotes). The metrics show '3 of 8' and 'Fixed 5 of 8 clean'. The recommendations say to do styles → structure → characters, then page, and the table shows what survives where if you do not fix: direct formatting into inconsistent weights, tabs into trim-dependent shifts, mixed quotes into wrong glyphs.
The same file after the style and structure pass. Word count still 75,000 at '.docx', but only direct formatting remains ticked. One issue at 6 minutes per 10k is 6 per block, '8 × 6 = 15 min' after the 15-minute floor. Verdict is 'warning' — "1 issue to fix before layout" — because one issue sits in the one-to-two warning band. The checklist now shows seven done and one not done. The metrics show '1 of 8' and 'Fixed 7 of 8 clean'. The recommendation still says to clear direct formatting before paginating, and the table now has seven rows where the PDF and EPUB cells say "will survive if not fixed" but the done flag is true, so the table is advisory.
A clean Google Docs export at 92,000 words. Word count 92,000 at 'gdoc', zero issues ticked. 'ceil(92000/10000)=10' blocks, '10 × 0 = 0' plus the 15-minute floor → '15 min' of final verify. Verdict is 'pass' — "File looks clean — final verify only" — because zero issues is the clean band. The checklist shows eight done, the metrics show '0 of 8' and '8 of 8 clean', and the recommendations say to run a final find for double space and double returns and to verify zero results in the exported PDF and EPUB.
What to verify in the export, not just in the source
The source pass is not the end of the check. The formatter must verify the export.
For print PDF, search for double space. The result should be zero. Double space that was fixed in the source can be reintroduced by the layout engine if the style carries a nominal space-after that the engine expands, but the source fix catches it before the engine sees it. Search the PDF, not just the '.docx'. For reflow EPUB, search for straight quote '"'. The result should be zero. Straight quotes that were normalized in the source can be reintroduced by pasted research that was not in the initial word count, so search the EPUB file itself.
For both exports, check the table of contents extraction. A dirty styles file extracts a dirty TOC — headings that were made by bolding text rather than applying Heading 1 do not appear in the TOC. The checklist's style pass is the TOC pass; after the map, extract the TOC and compare it to the expected chapter list. A missing chapter in the TOC is a direct formatting remnant.
When the file is not worth cleaning by hand
Some files are faster to rebuild than to clean. If the manuscript was written across machines with different smart quote settings, pasted heavily from email, and has direct formatting on every paragraph, the per-10k cost rises above the rebuild threshold. At 150,000 words with five issues, the estimate is '15 × 30 = 90 min' at the tool's 6-minutes-per-10k-per-issue rule, which is already a sitting. A file with extensive direct formatting on headings and body at 100,000 words can be 180 minutes of style mapping by hand, which is longer than applying the copy to a clean template. The tool caps the estimate at 240 minutes to signal that beyond four hours the economics favor a template apply, not another find/replace pass.
The check should run every time the file moves
The checklist is not a one-time gate. Every time the file moves — Google Docs to Word, Word to formatter, formatter back to author for new chapter — the same eight issues can reappear. Pasted text from email carries straight quotes; a new chapter typed quickly reintroduces manual tabs; a reviewer adds double spaces. Running the same checklist on every handoff costs 15 minutes of find/replace and avoids the case where a clean source becomes dirty on the last handoff before export. The tool reports 'Issues ticked 3 of 8' or '0 of 8' on every run so the author can see drift before the export hides it.
How long the pass really takes
The estimate is honest, not a marketing number. Six minutes per 10k words per issue is the measured rate for a careful find/replace with verification — find, confirm in context, replace, skip inside URLs or intentional double space after abbreviations. At that rate a 75,000-word file with three issues is 45 minutes. A 120,000-word file with four issues is 84 minutes, rounded to 85 and reported as '85 min' with the 5-minute granularity. The tool reports the estimate as 'Est. cleanup X min' so the author can budget a sitting and split at 120 minutes, which is where fatigue reintroduces issues.
Why the tool stops at the checklist
The tool does not rewrite the file. It names the issues, counts them, estimates the time, and tells you what survives where if you do not fix each one. The print PDF column and the EPUB column in the table exist so you can see that the same unchecked issue has two different survival modes — direct formatting into inconsistent weights in print and into missing styles in reflow, tabs into trim shift in print and into collapse in reflow. The same file has two fates; the checklist makes both visible before either export is paid for.
The free pass that is not free
Many formatters offer a "free cleanup" that is a pass of double spaces and double returns only. That pass catches two of eight. The remaining six — styles, tabs, quotes, dashes, pagination, and images — are the ones that generate the return trip when the proof shows uneven indents or wrong quotes. The checklist makes the full eight visible so the author's single pass before the formatter's pass catches all eight in the order that avoids rework.
A clean file is the cheapest edit you can buy
The manuscript cleanup is the least expensive edit in the production sequence — 15 to 90 minutes for most books — and it is the one edit that prevents every later cost from growing. A dirty file makes the typesetter slower, makes the EPUB engineer slower, makes the proof reader see phantom errors, and makes the formatter slower on the next change. A clean file makes each of those steps a mapping step. The checklist is a way to pay 45 minutes once so you do not pay it again as hours across the team.
A final worked pass from first save to final export
A debut author at 82,000 words in Word, three issues (direct formatting on chapter titles, tabs for indented dialogue, and a handful of hyphens where em dashes belong) starts with a duplicate. Save as 'manuscript-cleanup.docx' and open the styles pane. Create Normal at 12 on 15.5 Garamond with first-line 0.30 and space-after 6, create Heading 1 at 16 centred with space-before 18 and space-after 12, and create First Line Indent as a variant of Normal that guarantees the first-line rule. Select the chapter titles — there are 22 — and apply Heading 1 where the titles were previously made by selecting the text and pressing Bold and Center. Each apply removes the direct bold and center from the direct layer. After the 22 applies, select all body text and apply Normal, then reapply First Line Indent to the paragraphs that should indent and Block to the two block-quote passages so the direct indents are replaced. Clear the direct formatting cache with 'Clear All Formatting' on a sample and verify no stray remains.
Structure is next. Find '^t' (Word tab) and replace with empty, 82,000 words at 10 per line is about 8,200 lines, 22 chapters, roughly 180 tab characters found and removed. Find two spaces and replace with one — 47 doubles found from drafting speed, zero after second pass. Find '^p^p' (double return) and replace with '^p' — 18 empty paragraphs from scene breaks removed; reintroduce scene separation with 'space-after 12' on the style that precedes a break rather than an empty paragraph. Find ' - ' and replace with ' — ' where interruption is intended — 31 hyphens become 31 em dashes, two hyphens remain because they are in a range "pages 12-24" where en dash is correct. The pass is still 82,000 words; the word count does not change when tabs and doubles are removed, but the style count does — 22 Heading 1, 1 Normal, 1 First Line, 1 Block.
Characters are last in the middle. Enable smart quotes, then find straight '"' — 14 found in dialogue pasted from email, replaced after confirming pair direction. Find straight ''' — 6 found in contractions pasted from notes, replaced. Find '---' where the author typed three hyphens for an em dash — 4 found, replaced. Each find is a confirm-then-replace, not a blind all, because '"' can appear legitimately in a URL and ''' appears in a possessive that may already be curly.
Pagination is last. Delete the typed "Page 1" footer the author added in chapter 1 — one paragraph — and delete the typed chapter number header in chapter 5 where the author repeated the chapter title as a header line. Both become unnecessary when the engine paginates after the style pass.
The second run of the tool on the same file at 82,000 words now reports 0 of 8, 'pass', '15 min' final verify. Export the PDF and search for double space — zero. Extract the EPUB and search for straight '"' — zero. Extract the TOC from the exported PDF — 22 entries, one per Heading 1, where the dirty file extracted 11 because only half the titles had Heading 1. The dirty file's TOC miss is the most visible evidence that the style pass captured what direct formatting could not.