This is a distressingly common misperception. In fact the flux stream that you get out of a reader like a Greaseweazle is the TTL-level signal from the drive. This is the output of a differential amplifier and crossing detector. In theory, the raw drive head will show zero volts when nothing is changing under the read head, a very small positive voltage when the magnetic flux reverses in one direction, and a symmetric negative signal when it shifts the other way.
What you get out of the amplifier is a digital (!) pulse of ~1us when the signal shifts in either direction. And due to the way the amplifier works, when "nothing is changing" it tends to just amplify noise in the environment and invent fictitious transitions. Proper disk encodings limit the distance between transitions (generally to no more than three bit cells) to deal with this.
This behavior was actually a core copy protection trick. The custom drive writing the commercial software would gate the transitions in the middle of a sector, leading to a region that would read different noise every time. Correct consumer hardware isn't able to do that.
And that's why the flux files are so large: they're storing multiple revolutions of every track, precisely so that tools like the WOZ (Apple) or ATX (Atari) converters can intuit the style of copy protection (there seems to be no clean equivalent in the Commodore world, their "archived" copy protected media tends to be reverse engineered the last I checked?)
A fun project I'd love to see someone try some day would be to wire up a proper analog ADC chain (doesn't need to be more than ~8 bit at 2 Mhz or so) to the unamplified drive head output and see if a cleaner representation of the flux can be read.