I think I see what you mean, and if so I agree with the feeling, but what you actually wrote seems too strict. for example, if one were to use a proprietary text editor to make a change to the program (that's part of the workflow, right?), one may have just as much assurance that the modified program does what it's intended to do by verifying the change with a free diff tool. and the use of the proprietary text editor, though undesirable, doesn't render the program nonfree.

now, using a translation tool whose output is not practically verifiable (transpiler, compiler, assembler etc) may affect the freedom status of the output in the ways you describe and others. but the oppressor is not that who controls the compiled program, but that who controls the translation tool.

see, I'm not disputing the oppression, only attempting to frame it so that the response is directed to the oppressing party instead of to the victim