Semantic Nondeterminism — PDF Is the Proof | PQ PDF
No ads. No tracking. No data sold. Ever.
Switch tool
🏠 PDF Home
Core Manipulation<br>🗜️ Compress PDF<br>🗑️ Delete Pages<br>📐 Auto-crop & Deskew<br>📑 Extract Pages<br>📋 Flatten PDF<br>🎨 Grayscale / B&W<br>📎 Merge PDFs<br>📋 N-up / Imposition<br>🔧 Repair PDF<br>🔀 Reorder Pages<br>🔄 Rotate Pages<br>✂️ Split PDF
Convert<br>📝 PDF → Word<br>📊 PDF → Excel<br>🌐 PDF → HTML<br>🖼️ PDF → Images<br>📄 PDF → Markdown<br>📽️ PDF → PowerPoint<br>🗂️ PDF → PDF/A<br>🖨️ PDF → PDF/X<br>📊 Excel → PDF<br>🌐 HTML → PDF<br>🖼️ Images → PDF<br>📽️ PowerPoint → PDF<br>📝 Word → PDF
Forensics
🔬 PDF Forensics Scanner
📖 Scanner Description & Guide
🗂️ Office Forensics Scanner<br>🔬 Universal File Forensics<br>🔬 File Fingerprint Comparator<br>📧 PDF Forensics for Outlook
Security Research<br>🧪 All Research →<br>📊 PDF Forensics at Scale<br>🗂️ The Epstein Files, Forensically<br>🌀 PDF Reality Drift<br>🪞 The Illusion of Immutability<br>⚖️ Parser Disagreement<br>🧩 PDF Semantic Determinism<br>🤖 PDF AI Ingestion Pipelines<br>📋 PDF Form Security
Security & Privacy<br>🛡️ Protect PDF<br>🔓 Unlock PDF<br>⬛ Redact PDF<br>💧 Watermark<br>✍️ Sign PDF & PAdES
🖊️ Send for E-Signature
📖 E-Signature Description & Guide
Content & Annotation<br>📷 PDF Scanner<br>♿ Accessibility Checker<br>🎨 Colour Inspector<br>🔍 Compare PDFs
✏️ Edit PDF
📖 Editor Description & Guide
📄 Extract Text<br>📝 Fill PDF Form<br>🔤 Font Inspector<br>🔎 OCR PDF<br>🔖 Outline / Bookmarks<br>ℹ️ PDF Info<br>📊 Tables to JSON
Automation<br>⚙️ Workflow Builder
🔬 Security Research<br>Immutable Bytes, Mutable Meaning
A PDF isn't one document — it's a rendering program over an object graph, and the parser,<br>renderer, signature validator, or AI pipeline each decides what it says.<br>This is the single entry point to one argument, built in layers and measured across<br>24,824 real PDFs in three separate corpora.
Document AI assumes semantic determinism. The PDF format never provided it.
Ground truth, retrieval correctness, reproducible evaluation, hallucination control — every one quietly assumes a<br>document means the same thing to every reader. For PDFs that assumption is false, and now that machines do the reading it is a<br>measurable failure , not a curiosity. You don't have to care about PDFs to care about that.
Immutable bytes do not guarantee immutable meaning.
Imagine an LLM trained on a document no human has ever seen . Nothing was hidden and nothing was hacked — the<br>PDF simply held several valid machine-readable realities, the ingestion pipeline picked one, and enshrined it as ground truth.<br>That is not a thought experiment; it is what the studies below measure.
PDF was engineered to guarantee one thing: visual fidelity — that a page looks the<br>same on every screen and printer. It never promised semantic determinism — that<br>every system reading the file extracts the same meaning from it. For thirty years that gap was<br>invisible, because humans read the pixels and the assumption “one parser, one truth” was never tested.
Machines changed the stakes. RAG knowledge bases, LLM training corpora, compliance pipelines and<br>legal-discovery systems now read the object graph , not the page — and the gap becomes a<br>security and correctness problem: forms signed while their value and their appearance disagree,<br>knowledge bases that silently enshrine the wrong reality, evidence that reads one way to a person<br>and another to a tool. This program names that property, measures how often it occurs, and shows<br>what it enables.
Why this surfaced now
AI ingestion at scale1993 – 2022Humans read the pagethe gap is invisible2023 →Machines read the structurethe gap becomes a risk The flaw isn't new — the readers are. For thirty years a human read the rendered page and the<br>gap between pixel and payload never mattered. Now machines read the structure at scale, and a latent property of the<br>format becomes an active security and correctness problem.
For thirty years we assumed a document had one meaning, because a human read the page.<br>Now machines read the structure — and many PDFs never had a single meaning at all.
One file, as many documents as there are readers
ONE PDFsame bytes · one hashvalid signature👁Human readerthe page as drawn⌨Text extractionremapped characters🤖AI / RAG ingestiontext no human sees✍Signature validatorreports VALID ✓♿Accessibility treea different order The same bytes, the same hash, the same valid signature — handed to five systems that each<br>return a different document . None is malfunctioning; the PDF format guarantees they'll agree on how the<br>page looks, never on what it says.
See the gap in one file
The claim sounds abstract until you watch a single file say two things at once.<br>None of the files below is corrupt. None trips a malware alert. Each is simply<br>a different document depending on who — or what — is reading it.
Exhibit A The character that isn't the glyph
👁 What the page shows a human
9rendered on screen and in print
⌨ What text extraction returns
1copied, indexed, fed to the model
The font's ToUnicode map points the glyph...