Elysia Tools
Navigation
Workflow Playbook
Debug hard PDF extraction cases by checking encryption, safety, page scope, reading order, OCR, tables, review marks, and structured output.
Hubs
Difficult PDFs fail for different reasons. A password-protected file, a scanned contract, a tagged report with broken reading order, and a chart-heavy research appendix need different handling. This workflow keeps those causes separate so the team does not keep rerunning the wrong converter.
Use encrypted-pdf-converter to confirm whether permissions or password handling are part of the failure. Then run pdf-prompt-injection-scanner before the document is passed to AI-assisted summarization or workflow automation. That early screen creates a safer boundary between document evidence and instructions that should not be followed.
Most extraction bugs become clearer after narrowing the scope. Use pdf-page-range-extractor to create a small fixture and pdf-header-footer-noise-remover to reduce repeated boilerplate. Then inspect the file with tagged-pdf-inspector, pdf-reading-order-debugger, and formula-chart-heavy-pdf-analyzer to identify whether the main issue is tagging, reading order, scanned content, tables, formulas, or visual material.
Use scanned-pdf-ocr-to-markdown for image-only pages, pdf-to-structured-markdown-converter for readable prose, pdf-table-extractor-to-csv-json for tabular evidence, and pdf-to-json-structure-explorer when a hierarchy must be preserved. If reviewer deletions matter, add pdf-strikethrough-review-extractor before the final handoff. A good final note states which pages were tested, which output is trusted, and which content still needs human review.
Workflow playbook
Start by checking whether encryption, permissions, or embedded prompt-injection text could block or contaminate extraction.
Create a focused page subset and remove repeated headers or footers so debugging is not distorted by boilerplate.
Review tags, reading order, formulas, charts, and other layout-heavy regions to decide whether text extraction, OCR, or manual review is required.
Use OCR for scanned pages, structured markdown for prose, table extraction for grids, and JSON exploration when the result must preserve hierarchy.
Extract strikethrough and review signals, then document which pages, structures, and outputs are reliable enough for handoff.
Check encryption and prompt-injection content before sending the file into OCR, AI summarization, or automated downstream processing.
Extract a representative page range and compare tagged structure, reading order, headers, footers, formulas, charts, and scanned regions.
Choose OCR markdown for scanned pages, structured markdown for narrative text, table extraction for grids, and JSON exploration when a machine-readable hierarchy is required.