Different Rules/Scoring for Practice Sessions
I can understand the concept of retrying the lesson after three mistakes when I am learning a new skill. However I don't think this works well for the skill practice.
It feels particularly frustrating and time-wasteful when I fail a practice session by making mistakes outside of the lesson's scope.
So to be constructive, here are a few ideas for possible A/B experiments:
Don't fail practice sessions, but score them proportional to the number of the right answers.
Remove hearts only for mistakes in words that belong to the lesson. Of course you should still show the mistake.
Mark words that I got right as refreshed even if I fail the practice session.
This touches on how points are given for timed vs untimed practice. If I get 3 sentences right in timed practice, I get 3 points. If I get 3 sentences right in untimed practice, I get 0 points. This means that timed practice pays off much better in terms of points. Maybe Duo should even this out a bit. I definitely agree with your #3, maybe with #1 - after all, that is what happens if you do a timed practice session. (And I don't agree with #2, because that seems impractical to me, being a web developer myself this seems like a complex feature to maintain).
I agree, something should be done to encourage people do more untimed practice because it is more thoughtful and more useful when you are not very good at the current skill.
I usually did some lessons, then some untimed practice to catch most of the possible sentences (and read the discussions while nothing rushes me forward), then some timed practice to train reaction and fluency. However, rather often I moved to timed practice earlier because I hated to fail on untimed practice. This system gave me more points but, I think, less knowledge.
I'd like to add some thoughts about the suggested points.
Practice without failing is great. I also support the idea of continuous practice without any fixed number of sentences and phrases. The possibility of failure in a lesson is important because you should only move to the next lesson after you pass the current lesson with a good score. The idea of failing in a practice session has no ground in it: practice does not advance you down the tree but just improves your words' strength and gives you points.
If you can't fail in a practice session as is suggested in (1), you won't have any hearts at all, so this suggestion is not relevant. If there are still hearts, I don't agree with this suggestion: hearts should be lost for any mistake.
Again, if there is no failure in practice, this is not relevant. If there is a risk of failure, yes, I agree with this.
I didn't mean to put any priorities or dependencies there. E.g. having #2 doesn't assume implementation of #1. I merely suggest 3 different approaches that would solve my problem with the practice sessions.
I think Duolingo may want to run as experiment all three of them and then choose the approach that works best.
Hey ... who likes losing - but its a game. Some you lose and some you win!
That having been said, there are different strengths to be gained from both the timed and untimed practices, and from a points viewpoint, timed practice is more generous. This means that most of us (like me) do timed practice rather than untimed practice and lose out on the benefit that it offers. It stands to reason ...one can amass more points faster that way, and every right answer scores. So I go with (1) definitely. I don't feel strongly about (3) either way as I don't mind the extra practice, but I don't go for (2) - if the sentence is wrong - then its wrong and there should be no points or other benefits - end of story.
I couldn't care less about the points. The important thing for me is to spend my time only repeating things that I am about to forget. This the most efficient way of learning there is. However, currently repeating practices after failing 3 out of 18 makes me redo sentences that I have already got right. I perceive this is as a waste of my time.