Minification

Whitespace and line breaks, which the language does not need.

Comments, which describe intent to a reader.

Local identifier names, shortened where scoping proves it is safe. Names visible outside the module are preserved because renaming them would break callers.

Some tools also remove code proven unreachable, which overlaps with what a bundler does at a different stage.

It does not change behaviour, and a minifier that does is a bug rather than a trade.

Compression already handles repeated whitespace and repeated identifiers extremely well, since that is exactly what a compressor is good at.

So minification before compression saves meaningfully less than the raw size difference suggests.

The remaining gain is real and modest, and the parse-time gain from shorter identifiers is a genuine second benefit.

Which puts it firmly behind shipping less code and compressing what remains, in order of value.

It costs nothing to enable, so the ordering is about expectations rather than about whether to do it.

Every modern build tool minifies in its production mode by default.

Finding unminified output therefore usually means the build ran in development mode, which also means development warnings, assertions and often source maps are shipped.

That is a larger problem than the bytes: development builds of frameworks are substantially heavier and slower by design.

So the finding is worth treating as a build configuration question rather than as a minification question.

Checking the bundle for development-only strings settles it quickly.

Minified code is hard to debug, which is what source maps exist to restore.

Serving those maps publicly makes the original code readable again, which is a deliberate trade rather than a defect.

Uploading them to an error-reporting service instead keeps readable stack traces without publishing the source.

A map embedding the original source text is more revealing than one that only records positions, and both are ordinary build outputs.

JavaScript and CSS payloads are measured, and the heaviest individual files listed.

Whether text resources are compressed is read from the response, which is the larger of the two savings.

Publicly served source maps are reported separately, along with whether they embed the original code.

Whether a file is minified is apparent from its shape rather than being a declared property, so it is an inference from what was served.