Without looking at the code, direct observation does appear to agree with my understanding of the session summary vs. running tally (thumbs-up number):
I got 3+7+40 = 50 items correct out of 64 total items (4+11+49). 50/64 = 78.125%.
The session review definitely appears to be counting “items” (where an item is both the reading and the meaning for kanji and vocabulary, or just the name of radicals) and not individual, independent questions.
The running tally by the thumbs-up icon on each review screen, on the other hand, appears to be counting individual questions.
If I get, say, four meaning questions in a row (for four different kanji) and answer all but one correctly, the thumbs-up number reported is 75% (but if I quit at this point, the review summary would say 0% for 0 out of 0 answered since I hadn’t yet completed any “items”).
The meaning question and the reading question for a given kanji appears to count as a single item when computing the accuracy stat in the session review summary, but as TWO independent questions when calculating the running tally of the accuracy on each individual review page. [Potentially more than two questions if you miss either the reading or the reading one or more times.]
Sorry to prolong this thread, but I’m happy to have finally understood this: Session review summaries count “items” (radical names, or two reading+meaning questions for kanji and vocabulary). The thumbs-up running tally just counts independent questions and ignores whether they are for the same kanji or vocabulary item.
[UPDATE: I finally understand and appreciate everything being reported to me in the upper right of the individual review pages:
Thumbs-up: Percentage of correct answers so far. This is my bell-weather: if this drops much below 85% after doing a few dozen reviews, it becomes a grind.
Check-mark: The total number of “items” which you’ve answered both parts correctly: reading and pronunciation. It doesn’t matter if you miss one or the other multiple times, once you’ve eventually answered both parts of an item correctly, the check-mark count increases. Radicals only have one part: the name of the radical.
Inbox: This is the number of items remaining in your review queue.
I just finished a fairly difficult review of 118 items. On the final review page, the thumbs-up running total was 87%. It felt like an unusually tough session, but not the worst I’ve had.
The review summary only showed 78% accuracy though, because while I got 87% of the individual questions right, some of the correct answers were for questions I’d previously missed.
The review summary only counts an item as correct if you answer both halves correctly on the first try.
Whew. Back to your regularly scheduled programming!]