Book Isle: Audit suspected scan errors against page images and searchable text

A scanned PDF may contain both page images and searchable text, and they may not agree character for character. A name missing from search may reflect recognition or simply another spelling in the original. Use Book Isle for location and reading, with external notes for discrepancies. This claims no OCR generation, text-layer repair or original-image editing in the app. Work only with files you are authorized to handle and retain an unchanged original.

Prepare a traceable scanned sample

Choose a short scanned PDF you may read, preferably with clear page numbers and ordinary prose. Do not experiment with private identity documents or sensitive files. Record title, edition, source, filename, total pages and app version, then select one suspected error. For a self-made sample, write Lin borrowed a paper hat on paper, scan it and keep the paper original. Any searchable text is added by an external tool and must be identified in the record. A PDF extension does not establish a text layer, and import into Book Isle must not be assumed to recognize scanned text automatically.

Establish the page-image reference first

Open the PDF normally and record both file position and printed page number. Preserve both if they differ, for example file page eight and printed page five. Use a short passage before and after the target sentence as anchors, noting the column or caption. Inspect the whole page for reading direction and columns before joining text into a sentence. Zoom if needed, but do not retain only an isolated character without context. If the image is too blurred to identify, conclude that the original image is unconfirmed, not that recognized text is certainly wrong. Reliable checking starts by establishing the compared source.

Use search as evidence about available text

In a PDF-search entry actually available in the installed version, first search a clear, common word near the target, then the word containing the suspected error. Record query, hit status, location and displayed excerpt. If there is no entry or neither search returns a result, write that this path supplied no searchable text, not that the sentence is absent from the image. Use selection or copying only where actually offered; an image-only PDF need not support them. If necessary, examine the file’s text layer in an authorized independent PDF tool, clearly identifying that as an external result.

Keep three text sources in separate columns

In external notes, separate three columns: wording visible in the image, text obtained from the app or an external tool, and your proofreading transcription. For example, if the image clearly reads 纸帽 (zhǐ mào, paper hat), while a search excerpt reads 纸幅 (zhǐ fú, paper sheet), preserve both wordings and note the suspected substitution of 幅 (fú) for 帽 (mào) in the third column. Do not overwrite obtained text and make it appear that the file was correct originally. Preserve historical characters, variant forms and author spelling rather than silently modernizing them. If only 纸 (paper) is legible and the next character is unclear, retain a question mark in the transcription too. Fluent language is not a substitute for visual evidence.

Check details that can change meaning

Recheck numbers, decimal points, minus signs, names and negatives individually against the image, not just awkward wording. A missing not can reverse meaning, and a zero resembling a letter can change an identifier. These are risk types, not claims about your file. Attach page, line or column and context to every discrepancy instead of labelling the whole paragraph recognition failure. Another occurrence can provide supporting comparison, but cannot automatically replace the earlier wording. Distinguish a printing error in the original from a recognition error: their evidence and correction responsibility differ.

Use a separate external copy for any repair

If searchable text genuinely needs improvement, retain the original and save a revised copy in an external PDF tool you know and are authorized to use. Record tool, date and processing scope. This lesson neither requires a paid feature nor presents an external repair control as a Book Isle button. Before importing a clearly named revised copy normally, verify that page images remain, page count is unchanged and target text agrees with neighbouring prose. Do not overwrite the sole original or assume bookmarks and notes migrate losslessly. Record old and new entry identities separately; external proofreading notes may be sufficient without importing a replacement.

Narrow the investigation when results disagree

If an external tool and Book Isle search disagree, first confirm file revision, query and page position, then record both results without immediately diagnosing an indexing defect. If neither the image nor either text source resolves the wording, defer correction and seek a better authorized source or paper original. Do not upload an entire book to an unfamiliar recognition site for one character or send scans you cannot share. For reproducible behaviour, report version, query, position and visible difference using a self-made minimal sample. If only the large book is affected, state that limit rather than claiming all scanned PDFs are unsearchable.

Verify the evidence chain after leaving and returning

Leave the reader and return, relocating the discrepancy by file position, printed number and contextual anchors. Check that the three columns remain distinct. If you made a revised copy, open original and revised entries separately, confirming identity and target page. Classify the outcome as confirmed recognition discrepancy, possible original-source issue or unresolved; a suggested transcription is not a modified file. Completion means every proofreading change can be traced to visible evidence. Search assists location, while image, obtained text and personal judgement remain separately preserved.

Related guides and sources

Read PDFs in Book Isle: find information with pages, contents and search; Book Isle text search has no results: check scope, wording and extractable text.

Checked on November 7, 2026: Apple public version information lists Book Isle 1.1.4, requiring iOS / iPadOS 15.6 or later. Feature scope follows the official App Store description and the installed version. This article provides reading and checking methods you can perform; it does not claim step-by-step physical-device or private-server testing. Manual records, external tools and actually visible controls are identified separately, and personal analysis or code observations are not presented as verified automatic capabilities.

For the distinction between a scanned image and searchable text, see Adobe: Hidden OCR Text. This describes Adobe tools and does not establish equivalent editing functions in Book Isle.