Source maps eliminate the need to serve these separate files. Instead, a special source map file can be read by the browser to help it understand how to unpack your assets. It "maps" the current, modified asset to its "source" so you can view the source when debugging. This way you can serve assets in development in the exact same way as in production.

For example, imagine you are annotating the second page of a New York Times article. You probably want to see your annotation when you are looking at the article later as a single page, right? Or perhaps you've annotated the HTML for a PLOS ONE article. Wouldn't you like to see those annotations when you are looking at the PDF version of the same article? If annotations were only associated with the URL you happened to be looking at in your browser then the scenarios above would not work, because the documents being annotated all have different URLs.