RenderDiff
Evidence-first representation-divergence analysis for text, documents, and HTML. Find material differences between what humans see and what software actually receives.
What it does
RenderDiff detects representation divergence: situations where the apparent human-facing content differs materially from the bytes, Unicode, parsed structure, rendered state, or machine-facing representation actually processed by software.
Its central question is: What does the human think this says, what does the machine actually receive, and is the difference material?
Representation divergence can occur through invisible Unicode characters, homoglyphs and confusables, bidirectional controls, normalization differences, Unicode Tags, hidden HTML/CSS elements, ASCII smuggling, document extraction differences, and model or tokenizer boundaries. RenderDiff preserves the original evidence and identifies exactly where representations diverge instead of inferring intent from the existence of a difference.
Capabilities include original-byte preservation, codepoint inventory, NFC/NFKC comparison, Unicode 17 UTS #39 confusables analysis, invisible-character detection, bidi inspection, Unicode Tag decoding, HTML visibility analysis, optional networkless browser observation, tokenizer comparison, PDF extraction, selected Office-document extraction, deterministic JSON receipts, evidence bundles, optional Ed25519 signatures, SARIF output, HTML reports, PDF reports, CLI operation, repository scanning, and local web workflows.
RenderDiff is not a Unicode blacklist, generic malware scanner, AI-watermark attribution system, or universal safety certificate. Current public release: v0.6.1 beta.
Who it helps
Security engineers, application developers, document-processing teams, AI developers, trust-and-safety teams, researchers, and organizations dealing with multiple representations of the same content.