% correct seems to decrease with repeated errors on the same card

So I haven’t been able to confirm this exactly, but I suspect that answering incorrectly 2+ times on the same card x answer type decreases the %correct. Is this intentional?

1 Like

Yes, this is intentional. The % correct tracks the number of incorrect answer attempts against total answer attempts.

I could see an argument for tracking only a binary state of whether you got the item right or wrong though :thinking:

1 Like

That would make more sense to me. With the current system, theoretically, if I get all items wrong I would still get 50%, wouldn’t I?


Yeah I dont see the point in tracking multiple fails on the same character, it doesnt really have anything to so with SRS grades or even learning, mostly a “did you bother to learn from the mistake”.

1 Like

Yes, but just to make sure we’re on the same page that’s if you get all items wrong only once and then answer them all correctly.


E.g. if we have 10 item questions and we get all of them wrong once and then answer them correctly, then we would have 10 incorrect attempts over 20 total attempts, resulting in 50%. This percentage reflects the percent incorrect. Since we want the percent correct, we’ll subtract from 1. 1 - .50 = .50

If we add another incorrect answer attempt, we would now have 11 incorrect attempts over 21 total attempts. 11/21 ≈ .52 percent incorrect. To get the percent correct, 1 - .52 = .48

I guess the main point is that it gives you a rough estimate of how well you’re doing on the review session as a whole. But I can see how tracking multiple incorrect answers to be discouraging.

I’ll think about an alternative metric we could display :thinking:

Yeah I think it is pretty discouraging. It feels unfair. Like if I get 9/10 correct but I get one wrong 3 times, I’m at 9/12 = 75%? I don’t think that’s a good metric. Actually, I think you should consider showing the cards passed/total, as that’s what’s actually tied to SRS progression. Maybe have an on hover stats for nerds with other breakdowns such as reading %correct or meaning %correct.

1 Like

I actually had no idea it worked like that. I would’ve assumed that either my answer is right (increase SRS level) or wrong (decrease SRS level, but no more than once per session even if I get it wrong multiple times). So in that way binary accuracy makes the most sense.

This is assuming that once the card has been revealed to you you’ve already failed, and any additional attempts fall under ”re-learning”, similar to how a learning session can’t really punish you beyond the binary scenario.

1 Like