In image processing, the distinction between lossy (JPEG) and lossless (PNG) is visually apparent. In document engineering, "loss" often manifests as structural degradation rather than visual artifacting.
Vector to Raster Degradation
The most common lossy conversion is flattening vector data (text, SVG) into raster images. A document converted this way might visually resemble the original, but the text is no longer machine-readable, selectable, or indexable without Optical Character Recognition (OCR).
Semantic Loss
Converting a structured HTML document containing <h1>, <table>, and <em> tags into a basic PDF often results in semantic loss. The visual styling (bold, large font) is retained via absolute coordinates, but the structural meaning (this is a heading, this is tabular data) is destroyed unless the engine explicitly writes a Tagged PDF.
Mathematical Bounds of Conversion
When migrating from a high-entropy format (complex nested XML) to a low-entropy presentation layer, data must be discarded. The goal of an engineering-grade conversion pipeline is to control what is discarded and log the variance.