Duolingo treats humans like supervised large language models and only does the babbler stage. That is: they show you a bunch of data and labels and then do a loss function on that.
The problem is that humans aren't machine learning algorithms. Because of this, we need more. If you're an Latin based speaker, try Chinese, Japanese, or even Korean (especially Korean). If you haven't grown up around these characters you're going to really struggle with the introductory lessons. How do you resolve this? You get a book. ̶O̶n̶e̶ ̶t̶h̶a̶t̶ ̶t̶e̶a̶c̶h̶e̶s̶ ̶y̶o̶u̶ ̶t̶h̶e̶ ̶c̶h̶a̶r̶a̶c̶t̶e̶r̶s̶,̶ ̶h̶o̶w̶ ̶t̶h̶e̶y̶'̶r̶e̶ ̶f̶o̶r̶m̶e̶d̶,̶ ̶a̶n̶d̶ ̶h̶o̶w̶ ̶t̶o̶ ̶w̶r̶i̶t̶e̶ ̶t̶h̶e̶m̶ ̶(̶s̶o̶m̶e̶t̶h̶i̶n̶g̶ ̶D̶u̶o̶l̶i̶n̶g̶o̶ ̶d̶o̶e̶s̶n̶'̶t̶ ̶d̶o̶)̶.̶(see edit) The same is for grammar and even words.
It's amazing to me that in all this time Duolingo hasn't added a "supplementary materials" section. This is included in almost every single other app that's directed at these languages. I know they want to upset the market and do things differently, but that doesn't mean to throw the baby out too.
Duolingo is great for practicing and keeping motivated. It has successfully gamifyed language learning, getting users to practice frequently. I just wish they would put more focus on language learning, because it's common knowledge that Duolingo can't get you to even a conversational level. A years worth of studying shouldn't have that poor of a result.
Edit: it was wrong to claim that there isn't a letter learning lesson. There is a different tab for it and I made this claim off of prior experience. But the app drops you right into words without even knowing how words are constructed, and I think this is a grave mistake (I'm sure they have data that could test this: if there is a higher rate of struggling/dropouts when a new character system is introduced). I do not think this changes my thesis, but I want to call out my mistake. I'll have a followup comment in the replies.