Start the whole-book test with one passing chapter
Exporting every chapter at once makes a failure difficult to localize. The text, image reference, chapter selection, and reading order all become suspects at the same time. A staged workflow first proves one chapter with one image, then expands the manuscript to five chapters without changing the reference rule.
The general sequence is to define the minimum unit, record its starting values, plan and inspect the minimum EPUB, add chapters under the same conventions, and inspect the expanded EPUB with the same measurements. Passing the minimum version does not certify the finished book, but it gives every later failure a known return point.
Set a 1,200-word first-chapter target and one image
The English planning sample starts with a first chapter of about 1,200 words and one image. The planned asset is scene01.png, referenced with alternative text. In the dedicated Stage 4 copy, the corresponding test resource was recorded as figure01.png. The stable minimum is one chapter and one image, regardless of that fixture-name difference.
The Stage 4 result values report chapter count, image count, spine, missing resources, and structural validity. They do not report a fresh word count. The 1,200-word figure is therefore a planning target, not a measured result from the export evidence. If an exact word count matters, record it separately in the manuscript workflow.
Plan, export, and inspect the minimum version
The first plan selected one file and reported chapterCount 1. The minimum export was 4,982 bytes, and its recorded SHA-256 begins with 7d12c10d.
Inspection reported entryCount 10, spineItemCount 4, imageCount 1, zero missing resources, nav.xhtml, NCX, one item in each navigation representation, valid true, and no issues. The starting values—one chapter, one image, and one navigation item—therefore became the baseline for expansion.
If the image is missing at this stage, do not add more chapters. Return to the first chapter’s reference and the source image. If the reading order is wrong, return to selection and planning. The point of the minimum version is to keep the fault inside one chapter.
Expand to five chapters without changing the image-reference rule
After the minimum version passed, the manuscript was expanded to five chapters. The image folder and relative-reference convention were kept unchanged. The expanded plan reported fileCount 5, chapterCount 5, and valid true.
The expanded EPUB was 6,774 bytes, and its recorded SHA-256 begins with c4b96d. Inspection reported entryCount 14, spineItemCount 8, imageCount 1, zero missing resources, nav.xhtml, NCX, five items in each navigation representation, valid true, and no issues. The chapter count grew from one to five while the single packaged image and zero-missing result remained stable.
The measured change was chapters 1→5, spine 4→8, and navigation 1→5
Comparing the two artifacts shows chapterCount increasing from 1 to 5, spineItemCount from 4 to 8, and both navigation counts from 1 to 5. imageCount stayed at 1, and the missing count stayed at 0. The supported conclusion is that the controlled copy was expanded under the same image-reference pattern and the five-chapter local package remained structurally valid.
A separate common baseline used three chapters, one image, horizontal layout, and version 1.7. It reported seven spine items, zero missing resources, nav.xhtml, NCX, and valid true. That run supports the normal export route. The evidence for the one-to-five expansion is the paired title-specific artifacts.
Keep untested visual and external behavior outside the pass
The passed status applies to explicit exports of the one-chapter and five-chapter dedicated copies. It covers the inspected structure, reading order, navigation, registered image, and missing-resource result. It does not assess image appearance, how alternative text is presented, image rights, retailer recompression, KDP acceptance, hardware readers, or every reading application.
If the five-chapter version fails, return to the passing one-chapter baseline and add the remaining four chapters in smaller groups. If the image looks wrong, return to the source asset. If an external service reports a problem, use that environment’s diagnostic. Local valid true should not be expanded into a universal quality guarantee.
Put Rune Studio inside the staged workflow
Rune Studio publicly supports generating EPUB 3 from Markdown manuscripts. Its export process registers chapters in the spine, creates nav.xhtml and NCX navigation, and packages referenced images. The package document, manifest, and spine roles are defined by the EPUB 3.3 specification.
This run applied that public capability explicitly to a dedicated copy. It did not test automatic tracking of Finder or cloud-storage changes, retailer acceptance, or device rendering. Rune Studio is the controlled export and inspection route in this staged method, not evidence for those external outcomes.
Conclusion: preserve the one-chapter pass while expanding to five
A manageable image-based EPUB conversion begins with a roughly 1,200-word first-chapter target and one image, then verifies one chapter, one packaged image, zero missing resources, and valid structure. Only after that baseline passes should the same reference rule be extended to five chapters and checked again.
The dedicated sample completed both stages: one chapter became five, the spine grew from four items to eight, navigation grew from one item to five, and the one packaged image remained present with zero missing resources. If a later value changes unexpectedly, return to the passing one-chapter artifact and isolate the newly added chapters.
