Thanks for the feedback, @Doc James.
You brought good points. I'll provide a bit of context on why things are in the current way and aspects that we can adjust.
Regarding the direct access to the third step. We will support having direct access to the second step since the first step is useless when there is an exact search result for the page on the URL. As part of Section Translation we'll update the dashboard for Content Translation and this will be supported in the new dashboard.
However, it may be hard to skip the second step (confirmation step) completely to get to the third one. The reason is that Content Translation saves the user progress automatically (which is convenient to work in multiple sessions) and we need to show the terms under which content is stored before storing it. That is, the confirmation step is telling the user that their contributions are shared under a Creative Commons license and they need to see it before contributing. Alternative approaches where such info is shown before the first time saving occurs in the editor (when the user is focused on the translation task) are potentially distracting, but we are open for exploring other ideas.
Regarding adding all the contents to the translation, the tool was designed to provide freedom on how much content the user needs to translate to create an initial version of the article. Adding the whole article to the translation can add more pressure to the editor when dealing with long articles or complex contents that some users prefer to skip. For your particular case, articles have been pre-filtered already to include the contents that are expected to be translated, but often editors take those decisions as they translate, and our research suggests that it is useful to have visual feedback that helps to distinguish what has been translated and what is pending.
Technically, since we preload the translation for the next paragraph and users had to place their cursor anyways, the click users make on each paragraph to add it would be done anyways to place the cursor in order to edit. So I'd not expect impact on the process fluency while it provides the benefits mentioned above. Having said that, I'm happy to hear more details around this and explore how can provide better support.