I’m interested in this as well. We can probably figure it out if multiple people post some stats. To that end, I did the first round of 15 lessons today and the batching was as follows:
- L9 Vocab, L9 Vocab, L10 Kanji, L9 Vocab, L9 Vocab
- L9 Vocab, L9 Vocab, L10 Radical, L10 Kanji, L9 Vocab
- L9 Vocab, L9 Vocab, L10 Radical, L10 Kanji, L9 Vocab
The available lessons prior to starting them were as follows:
| Type | Amount |
|---|---|
| L9 Vocab | 83 |
| L10 Radicals | 15 |
| L10 Kanji | 23 |
| L10 Vocab | 42 |
EDIT: So, my initial guess based on very limited data thus far is that it’s based on percentages, prefers older vocabulary over new (thus ignoring the new vocab until all old vocab is cleared out), and tries to mix at least one of each type when possible. Based on the above totals and applying each percentage to the 15 lessons to do:
EDIT 2: Based on more data posted by @LupoMikti, I’ve updated this with a refined guess that works for both data sets. The only difference from the initial guess is that it probably takes the ceiling instead of rounding and then potentially subtracts off one or two random items from a random type if the calculated amounts sum to more than total number of daily lessons due to the rounding error.
EDIT 3: @tofugu-scott has now described the algorithm further down. It confirms the aforementioned refined guess and adds a few more details about the interleaving.
| Type | Amount | Percentage of Total (excluding Level 10 vocab) | Applied to 15 lessons, ceiling |
|---|---|---|---|
| L9 Vocab | 83 | ~68.6% | 11 |
| L10 Radicals | 15 | ~12.4% | 2 |
| L10 Kanji | 23 | ~19% | 3 |
That matches up exactly with the actual number of lessons of each that I was presented with allowing for removing one of the vocabulary since 11+2+3 = 16 which is clearly greater than the 15 total available slots.