App dependency matrix: "needs", "improves", "breaks"
← The Map · Table of contents
Summary
For every app, three questions: What does it need to have happened before? What does it improve for the apps that come after? What can it break if you run it at the wrong moment? This page is the quick-lookup version of the whole book's "Works well with" lists. First the famous chains, then the full matrix.
Details
Chains worth memorising
| Chain |
What is going on |
| Rotation → Searchable → Search |
Orientation errors hurt recognition most; Searchable adds the text layer; Search then finds words quickly without redoing recognition each time |
| Deskew → Searchable |
Deskew rebuilds pages as images (likely no text layer), so recognition goes after, never before |
| Merger → Searchable |
Photos need to become a PDF before they can get a text layer |
| Searchable → Extract Text |
Extract Text reads only existing text, so a scan gives nothing until it has been made searchable |
| Searchable → TOC AI / Ask PDF |
Both need real text |
| Unlock → almost everything |
Merger refuses encrypted files; Searchable, Compressor, TOC AI, Flatten, Converter and others expect an open file |
| Delete Page / Merger → Add TOC |
Contents are page numbers; change pages first, number them after |
| Extract Form Data → Flatten |
Read the answers before flattening turns the fields into static content |
| Flatten → Compressor → Protector → Sign |
The book's order for a document that ends up locked and certified (with a caveat below) |
| Compressor → Protector |
Shrink while the file is open; lock afterwards |
| Any edit → Sign → Verify Signature |
Sign last, check straight away |
| Recovery → everything |
A damaged file cannot be trusted to feed anything else |
The full matrix
Legend: Needs = what should be true of the input. Improves = which later apps or goals get better after this app. Can break = what it can damage. "Expected, not tested" marks an inference.
Organize
| App |
Needs |
Improves |
Can break |
| Merger |
Unlocked inputs; upright pages; files in the order you want |
One input for Add TOC, Compressor, Searchable and others; with OCR ticked, scans get text |
Page order (it orders files, not pages); encrypted inputs are refused; any signature on an input (expected, not tested) |
| Splitter |
Unlocked; page numbers noted |
Extracts for Extract Image, Checklist, sharing a part |
Bookmarks and links across pieces; total size can grow; signatures (expected, not tested) |
| Delete Page |
Unlocked; page numbers noted |
Smaller, cleaner file; sharing safely a whole-page subset |
Existing contents page and bookmarks may point to wrong pages; signatures (expected, not tested) |
| Rotation |
Know which pages are wrong |
Recognition quality for Searchable and OCR Image |
Auto Skew resamples content (text layer: expected, not tested); signatures (expected, not tested) |
| Resize |
Final content |
Printing, Booklet |
Tile mode converts pages to images (file grows; text only if OCR ticked); signatures (expected, not tested) |
| Booklet |
Final content at half-sheet size |
Print-ready output |
Reading copy: the result is a print layout, so do not use it as the document to share on screen |
Add & Structure
| App |
Needs |
Improves |
Can break |
| Add TOC |
Final page order |
Navigation; Merger packs; Abstract pages become listed |
Becomes wrong if pages change afterwards; signature (expected, not tested) |
| TOC AI |
Text layer; unlocked; final page order |
Navigation in long documents |
Outline goes stale if pages change later; may stop early on the free plan (continuation offered) |
| Add Table |
.xlsx; final page order |
Data appendices |
Shifts page numbers after the insert point, which stales any earlier contents |
| Flatten |
Unlocked; form filled |
Consistent display; clean input for Sign and PDF/A Validator |
Form data: Extract Form Data returns empty after it |
| Collage |
Photos |
Print sheet for Add TOC, Booklet |
Crops pictures (AI chooses framing) |
| AI Illustrator |
TXT/PDF with text; structure finished first |
A gift-quality book that TOC AI can then outline |
Not for business documents; output is PDF only |
Protect & Prove
| App |
Needs |
Improves |
Can break |
| Protector |
Final content; unlocked and unsigned input |
Confidential sharing |
Everything downstream: most apps cannot read the result; do not edit afterwards; a signature made before it is expected to break |
| Unlock |
The password |
Every app that needs an open file |
A digital signature on the file (can become invalid; verify on the live app) |
| Sign |
Final content; .pfx and password; normally no open-password |
Proof the file is unchanged since signing |
Nothing, but it ends editing: any change afterwards invalidates the signature |
| Verify Signature |
The file exactly as received |
Confidence after any step |
Nothing: read-only. Re-saving or compressing the file first would destroy what it tests |
| PDF/A Validator |
The final file, unlocked |
Evidence for archive submissions |
Nothing: read-only. But any rewrite after validating (for example Compressor) means validate again |
Shrink & Repair
| App |
Needs |
Improves |
Can break |
| Compressor |
All edits done; unlocked; healthy |
Email and upload limits; makes the file small after Searchable grew it |
Signatures; picture quality (lossy, do not run twice); PDF/A status (validate after) |
| Optimize |
A normal PDF |
Web-fast display; modest size savings; tidy input for the PDF/A check |
Little (lossless by the site's statement); rewrites the file, so signatures (expected, not tested) |
| Deskew |
Upright, scanned pages |
Recognition accuracy; archive quality |
Text layer (likely lost; verify); pages become re-encoded images |
| Recovery |
One damaged, unlocked file |
Brings a broken file back so other apps can use it |
Default "Full" policy removes metadata, attachments, history and hidden content; reconstructed pages can differ |
| Unwatermark |
Real text; one file |
Clean provenance for AI-generated text |
Layout (images, tables, columns, fonts, links lost); original metadata; rewriting can alter wording |
| App |
Needs |
Improves |
Can break |
| OCR Image |
Upright JPG/PNG; right language |
Quick text from pictures |
Nothing (read-only); output has no layout |
| Searchable |
Upright scanned PDF; unlocked; right language |
Search, Extract Text, TOC AI, Parser text, all AI-understand apps |
File size grows; AI correction can alter words; Word/Excel/PowerPoint output drops the page picture |
| Extract Text |
Text layer |
Feeding editors and other tools |
Nothing (read-only) |
| Extract Image |
Unlocked |
Rebuilding with Merger; OCR Image input |
Nothing (read-only) |
| Extract Metadata |
Any PDF |
Audits before publishing |
Nothing (read-only) |
| Extract Form Data |
Interactive filled form |
Spreadsheets of answers |
Nothing (read-only); empty on flattened forms |
| Parser |
Unlocked; text layer for the text file |
One-shot audit; backup of outline and form data before editing |
Nothing (read-only) |
| Search |
Text layer (or tick scanned and set language) |
Marked copy for colleagues; page numbers for Delete Page |
Not destructive; on scans, misrecognised words are missed |
| Global Search |
Nothing |
Finding a tool or article |
Nothing |
| Resume Analyzer |
A CV |
Better CVs; structured data |
Nothing to the file; scores are AI opinion |
Convert & Create
| App |
Needs |
Improves |
Can break |
| Converter |
Unlocked input; matching content type |
Brings odd formats into the suite; PDF/A production; editable Office copies |
Layout fidelity in either direction; treat PDF to Word as a draft |
| Chat |
Real text if you upload a PDF |
First drafts; small wording edits |
Accuracy (AI-written); scanned input gives it nothing to read |
| SVG Maker |
An idea |
Art for covers and diagrams |
Nothing downstream; the PDF copy is a picture, not editable vectors |
AI: understand
| App |
Needs |
Improves |
Can break |
| Ask PDF |
Text PDFs |
Fast, verifiable answers |
Nothing (read-only); stops with "looks scanned" when no text |
| Abstracts |
PDF |
A front-page summary |
Adds a page, so page numbers after the front shift (and contents entries: expected, not tested); run Add TOC or TOC AI after |
| Cheat Sheet |
Text PDF or a topic |
Study and onboarding sheets |
A scan plus a topic may silently use only the topic |
| Checklist |
Text PDF and a specific topic |
Action plans |
Scans give a placeholder checklist |
| PDF Knowledge Base |
Text PDFs |
Questions across a whole collection |
Expires in about a day; download the ZIP |
AI for developers
These apps do not touch documents, so they sit outside the document matrix. Their own dependencies:
| App |
Needs |
Improves |
Can break |
| Assembly Chat |
A .NET assembly or NuGet name |
Understanding before an upgrade |
Nothing |
| Code Graph |
GitHub repo with a .sln |
Feeds Repo Docs and Refactor Assistant thinking |
Nothing (repository not changed) |
| Repo Docs |
GitHub repo; access for private ones |
Onboarding |
Opens a pull request; review it |
| Refactor Assistant |
C# repo |
Smaller classes |
Apply opens a pull request; your CI is the build check |
| NuGet Updater |
SDK-style .NET repo |
Fresh dependencies |
Pull-request mode changes version values; check build |
| Page SEO Checker |
A public URL or sitemap |
Better titles and descriptions |
Nothing (CSV only) |
| Advisor and Document Engineer |
A request in words |
Picks the right Documentize tool; builds a utility |
Nothing (code is built, never run) |
The tension at the end: Protector versus Sign
The book's ordering rule for a document that is both locked and signed is edit → flatten → compress → protect → sign. But Sign works on a file without an open-password, and Protector rewrites the file, which would break an earlier signature. Whether a protected file is accepted by Sign: verify on the live app. If in doubt, sign first, then share the file by a protected channel, or protect a copy. This is the one place where the ideal order is genuinely unresolved.