iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
A RAG pipeline can fail on a real document even when its OCR looks accurate: text may be missing, reordered, detached from its headings or table, or split into chunks that no longer contain enough evidence to answer a question. Find the first stage where the document stops matching the source page—extraction, structure, chunking, retrieval, or answer generation—before tuning later stages.
Why do real documents break a RAG pipeline?
Retrieval-augmented generation (RAG) preprocesses and indexes source material, retrieves passages for a query, then gives those passages to a language model to help form an answer. Each step depends on the previous one preserving useful evidence. A clean text file may pass through this chain with little difficulty; a scanned PDF, a multi-column report, or a page with tables may not.
PDFs and images can require text conversion, OCR, or layout-aware parsing. Even when words are recognized, a plain-text extraction can flatten or reorder headings, lists, footnotes, and tables. A table’s values, for example, may survive while their row and column relationships do not. The UK Government’s RAG overview emphasizes that preprocessing depends on data type and that splitting method and chunk size affect performance and context relevance.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesReadable text is not necessarily usable evidence
Character- or word-level OCR accuracy measures whether recognized text resembles the source. It does not by itself show that a passage retains the relationships needed to retrieve and answer a question. An Association for Computational Linguistics (ACL) 2026 Industry Track paper covering 11 challenging document types warns that structural and semantic errors can cause substantial retrieval failures even when word error rate (WER) or character error rate (CER) is low. Its examples include complex layouts, watermarked backgrounds, historical pages with unusual reading order, decorated text, tables, and formulas. The finding is a warning about the limits of OCR scores, not a claim that every OCR system or corpus behaves the same way.
#1 Best Overall
- PORTABLE SCANNER FOR USE ON-THE-GO — The fastest and lightest mobile single-sheet-fed compact document scanner in its class¹
- QUICK DOCUMENT SCANNING ― This Epson ultra-fast scanner scans a single page as quickly as 5.5 seconds²; Windows and Mac compatible
- VERSATILE PAPER HANDLING ― Portable scanner scans documents up to 8.5 x 72 in; Also easily digitizes receipts and ID cards to make accounting, bookkeeping, and organizing simpler
- INTUITIVE, HIGH-SPEED SOFTWARE — Epson ScanSmart Software³ is a smart tool allowing you to easily scan, review, and save; Stay organized easily with the help of this Epson scanner
- EASY SETUP — USB-powered connect to your computer for quick and simple scanning; No batteries or external power supply required to operate portable document scanner; Standard Connectivity: USB 2.0
Chunking can discard context that extraction preserved
A passage may be extracted correctly but become ambiguous when split away from its heading, the question it answers, or the labels that explain a table. Very large chunks can also make retrieved context less focused. There is no universally best chunk size established by these sources: the appropriate split depends on the document structure and the questions the system must answer.
Google Cloud documents one layout-aware option: its parser can detect text blocks, tables, lists, titles, headings, headers, and footnotes in supported file types. Its chunking configuration can keep a chunk within a layout entity and append ancestor headings. These are capabilities of that service, not evidence that it is optimal for every corpus. Its documented configuration offers a 100–500-token option and a 500-token default; those figures apply to that service, not to RAG generally.
Rank #2
- FAST SPEEDS - Scans color and black and white documents a blazing speed up to 16ppm (1). Color scanning won’t slow you down as the color scan speed is the same as the black and white scan speed.
- ULTRA COMPACT – At less than 1 foot in length and only about 1. 5lbs in weight you can fit this device virtually anywhere (a bag, a purse, even a pocket).
- READY WHENEVER YOU ARE – The DS-640 mobile scanner is powered via an included micro USB 3. 0 cable allowing you to use it even where there is no outlet available. Plug it into you PC or laptop and you are ready to scan.
- WORKS YOUR WAY – Use the Brother free iPrint&Scan desktop app for scanning to multiple “Scan-to” destinations like PC, Network, cloud services, Email and OCR. (2) Supports Windows, Mac and Linux and TWAIN/WIA for PC/ICA for Mac/SANE drivers. (3)
- OPTIMIZE IMAGES AND TEXT – Automatic color detection/adjustment, image rotation (PC only), bleed through prevention/background removal, text enhancement, color drop to enhance scans. Software suite includes document management and OCR software. (4)
How can you locate the first broken stage?
Use a small set of questions with known answers and supporting page locations, then follow the evidence forward. The aim is to identify the earliest mismatch. If extracted content is already wrong, adjusting prompts cannot restore missing or misordered source information.
Free tools Windows power users keep installed
One-click scans. No signup required.
1. Build a representative test set
Collect documents that reflect what users actually query, including both straightforward examples and difficult ones such as scans, tables, multi-column pages, footnotes, and forms. Pair them with questions whose answers and supporting locations are known. Microsoft Learn’s Azure Architecture Center evaluation guidance starts with collecting test documents and queries before end-to-end evaluation.
Rank #3
- FAST DOCUMENT SCANNING — Document scanner with feeder allows you to speed through stacks with a 50-sheet Auto Document Feeder (ADF); Efficient office scanner to help you scan more productively
- INTUITIVE, HIGH-SPEED SOFTWARE — Quickly scan with this desktop document scanner; Epson ScanSmart Software lets you easily preview scans, email files, upload to the cloud, and more; Plus, automatic file naming saves even more time
- SEAMLESS INTEGRATION — Easily incorporate your data into most document management software with the included TWAIN driver; Office document scanner integrates seamlessly with business workflows
- EASY SHARING — Duplex scanner allows you to scan straight to email or popular cloud storage2 services like Dropbox, Evernote, Google Drive, and OneDrive for simple storage and sharing
- SIMPLE FILE MANAGEMENT — Scanner allows the creation of searchable PDFs with Optical Character Recognition (OCR) and convert scans to editable Word or Excel files effortlessly; Designed for home and office document scanning
2. Compare extraction with the source page
For each test page, check whether text is missing, reading order is wrong, table values have become detached from their labels, or headings and section boundaries have disappeared. For OCR-based pages, do not treat a low WER or CER as proof that the extracted material will work as RAG evidence.
3. Inspect the indexed chunks
Open the actual chunks stored for retrieval, rather than relying only on a preview of the parsed document. Check whether a passage retains the heading needed to interpret it, whether table context makes sense, whether a split has separated essential context, and whether source-location metadata survives. Google Cloud’s documentation describes appending ancestor headings as one way to reduce context loss in chunk retrieval and ranking.
Rank #4
- Scanner type: Document
- Connectivity technology: USB
- With Auto Scan Mode, the scanner automatically detects what you're scanning
- Digitize documents and images
4. Test retrieval without generation
Run each known question against the index and inspect the returned passages. Ask two separate questions: does the retrieved material contain the answer, and does it contain enough of the answer? If the answer is absent from the retrieved context, the failure is upstream of answer generation. If relevant passages are present but the answer is incomplete or unsupported, investigate how the model uses that context.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstall5. Score answers on distinct dimensions
Microsoft’s evaluation guidance identifies groundedness, completeness, utilization, relevance, and correctness as useful dimensions. Completeness asks whether all parts of a query are answered; utilization asks how much of the returned context is actually used. These measures help distinguish a missing-evidence problem from a failure to use available evidence. Microsoft also notes that language-model responses are nondeterministic, so evaluation targets are better considered as ranges than as a single guaranteed result.
Best Value
- OUR MOST ADVANCED SCANSNAP. Large touchscreen, fast 45ppm double-sided scanning, 100-sheet document feeder, Wi-Fi and USB connectivity, automatic optimizations, and support for cloud services. Upgraded replacement for the discontinued iX1600
- CUSTOMIZABLE. SHARABLE. Select personalized profiles from the touchscreen. Send to PC, Mac, mobile devices, and clouds. QUICK MENU lets you quickly scan-drag-drop to your favorite computer apps
- STABLE WIRELESS OR USB CONNECTION. Built-in Wi-Fi 6 for the fastest and most secure scanning. Connect to smart devices or cloud services without a computer. USB-C connection also available
- PHOTO AND DOCUMENT ORGANIZATION MADE EFFORTLESS. Easily manage, edit, and use scanned data from documents, receipts, photos, and business cards. Automatically optimize, name, and sort files
- AVOIDS PAPER JAMS AND DAMAGE. Features a brake roller system to feed paper smoothly, a multi-feed sensor that detects pages stuck together, and skew detection to prevent paper damage and data loss
6. Change one cause at a time
Use the observed failure to choose what to change: parsing for missing or reordered content, chunk boundaries for lost context, embeddings or retrieval settings for poor passage selection, or prompts for answer behavior. Re-test against the same representative questions after each change. In the UK Government’s described workflow, changing chunking or embeddings requires re-indexing all chunks; plan for that cost rather than assuming an index will update itself.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How should you interpret published benchmark numbers?
Benchmark results can show that document preparation choices matter, but they are not portable promises or a universal parser ranking. A 2026 arXiv preprint evaluated a manually curated corpus of 36 Portuguese administrative documents: 1,706 pages and approximately 492,000 words, tested with 50 questions. The authors reported the following scores using their evaluation method and LLM-as-judge results averaged over 10 runs:
| Configuration in the paper | Reported score | What the result represents |
|---|---|---|
| Naive PDFLoader baseline | 86.9% | Score on the paper’s Portuguese administrative-document corpus and evaluation method |
| Manually curated Markdown | 97.1% | Score on the same corpus and evaluation method |
| Best automated configuration | 94.1% | Score on the same corpus and evaluation method |
The authors also reported that metadata enrichment and hierarchy-aware chunking contributed more than conversion-framework choice alone. Treat these results as evidence about that test corpus, not as expected performance on another language, document mix, question set, or evaluation setup.
Recommended Free Tools
How do you choose what to improve?
Compare ingestion or retrieval approaches against the documents and questions your system must handle, not a single generic accuracy number. For each candidate, check:
- Document coverage: Does it handle your mix of scans, digital PDFs, tables, formulas, and layouts?
- Structure preservation: Does it retain reading order, hierarchy, table relationships, and source locations needed to retrieve and cite evidence?
- Retrieval usefulness: Do passages returned for representative questions contain enough relevant evidence, and are irrelevant passages also returned?
- Answer behavior: Are answers grounded, complete, relevant, and correct, and does the model use the supplied context?
- Change cost: Will changing chunking or embeddings require a full re-index? A managed service may also constrain settings or lock configuration when a data store is created.
The practical rule is to fix the earliest stage that fails inspection, then run the same questions through the pipeline again. Stronger generation cannot compensate for evidence that was lost or made uninterpretable before retrieval.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
