Lossless compression means that a decoder can reproduce exactly the data given to the compressor. PNG can recover the same pixel values; ZIP can recover the same bytes. The claim is precise and local. It says nothing about whether the input already lost information during an earlier conversion.
Where does the loss happen?
Imagine a spreadsheet with a formula, a displayed number, and a formatting rule. Export it to a screenshot, then compress that screenshot as PNG. PNG is lossless with respect to the pixels. The formula, cell relationship, and rule disappeared when the spreadsheet became an image. Calling the PNG lossless is correct; treating it as a lossless copy of the workbook is not.
Why can’t a perfect decoder recover the rest?
Several originals can produce the same visible result. A cell showing `12` might contain a typed number, `6+6`, or `3*4`. Once only the image remains, no decompressor can know which history occurred. A reconstruction system may infer a plausible formula from nearby labels, but inference is not evidence that the original formula has been recovered.
What should “lossless” be compared with?
The useful comparison is between two specific stages: original pixels and decoded pixels, original bytes and restored bytes, or a source document and its export. A WAV file and an MP3 make a different claim from a PNG and a screenshot. The word describes a transformation, not the entire history of an object.
What should an archive record?
It should record the source object, the conversion step, and the resulting format. That makes it possible to say exactly what was preserved and what was discarded. A lossless export can then serve its purpose without being mistaken for a complete copy of the original editing environment.
The distinction also matters when a format has optional metadata or multiple valid decoders. Exact bytes may return while timestamps, colour profiles, or application-specific fields remain outside the compressed payload. “Lossless” should always name the input and the boundary being tested.
The same distinction helps with audio, video, and documents. Ask what the next reader must preserve—samples, structure, timing, editability, metadata—and choose the loss boundary deliberately.