An experiment with phonetic writing

[rikforto]

I find the other possibility, that they were just taking down sounds, and badly, hard to credit.

Imagine that you are an Englishman who has come to France. You need to learn the language, and first and foremost, you need to learn to speak it, but not to write it. You have at least roughly learned the necessary words at a conversational level.
Now you will need to write down a book in French... You can’t read, so you hire someone to dictate it to you. Overall, you’re just recording sounds. And it’s very bad, because you don’t know how to record them correctly. 
The result will likely be a hard‑to‑read tirade, because, being untrained in written French, you will write the word aujourd’hui as ujuvi (or something similar).

In this thought experiment I am carrying on conversations in French! To do so I must be able to reliably distinguish between phonemes in context, both as a listener and to a reasonable degree produce them. Otherwise I cannot distinguish between words in a conversation and I cannot participate in those conversations. I cannot then sit down to transcribe the dictation and suddenly be robbed of my ability to make these determinations. I also decided how to correctly transcribe those distinctions---I do know how to record them correctly!

I am not saying there is no room for error, especially if the target text is harder than my conversational level. But if I cannot recover from "ujuvi" "aujourd'hui", I have not recorded French there because the entire purpose of the recording process is to facilitate recovery. If this is the case at scale, I have failed to transcribe French. I have a very hard believing that large tracts of the VMS are incomprehensible for this reason because I would know this as I wrote it because I can speak French and would know if what I was writing was incomprehensible to me! This strains credibility to me; what is this document even supposed to be for?

At any rate, the fact that people who have no exposure to the language of dictation cannot accurately transcribe it is a poor analogy to a thought experiment where the person is presumed to not only be a reasonably high-level speaker of the language, but to have designed an alphabet for systematic use.