There are a couple of reasons for this!
First, Japanese sentence parsers are adept at parsing sentences that contain kanji, but not quite as good when it’s all kana. This means Yotsuba’s dialogue has a higher error rate when parsing for words.
If you find your percent of known words for the series is highest in volume one, that’d likely be because Yotsuba uses kanji in the first volume.
The second reason is that Manga Kotoba tracks known words separately when they appear with kanji, as hiragana, and as katakana. There are probably a bunch of words you know that appear as hiragana in Yotsuba, but you only have them marked as known with kanji. (At least, that’s been my own experience.)
It is a bit misleading, but I don’t think there’s any good way around it, unfortunately!
One of my goals for Manga Kotoba is to create the site I wish I had when I started out reading.
The best part is even after you learn many words, and encounter many series where you know a large percentage of words, you can always discover new genres where you’re back at square one, and have a vast number of high frequency words to tackle.
This can be an issue sometimes. I learned to accept it as the price of learning from a frequency list. “Oh, this word means amusement park. I guess I know what’s coming up in this volume.”
But as a motivating factor, the more words you learn, the fewer potential for spoilers you’ll encounter in other series’ frequency lists 
I also agree about the lack of context, which softens the spoiler blow a bit as you have no idea how they’ll apply until you encounter them (and hopefully recognize them without having to stop and look them up!)
The best part is there’s still room for improvement.
For example, there’s a big difference between “I will need to look up one word per sentence” and “I will need to look up two or more words per sentence”. Not just the time spend looking up words, but also because the more words one has to look up, the easier it is to loose track of the meaning of the sentence. So once I add in percentage information for “+1 sentences” that require looking up only one word, it will make it clearer whether those unknown words will be a low or high hindrance when reading.
I struggled through the “gen one” Pokemon manga, and I can’t even put my finger on why. I think part of it was just an amazing amount of word lookups I had to do along the way. When I eventually pick it back up with the “gen two” volumes, I hope to have a better time of it as this time I’ll be able to pre-learn words.
It’s possible there are some character, Pokemon, and location names I haven’t added to the block list yet. It’s one of the disadvantages to frequency lists, and has a greater impact in series where I haven’t blocked any character names yet.
Every time I try to go the pre-made deck route (whether to use myself, or to create for others to use) I do end up with more negatives than positives.
Sites like JPDB also have the disadvantage of misparings that end up in decks.
Manga Kotoba falls into this category, but has it even worse as the manga OCR process can misread a kanji, and that’s what ends up on the site (or else gets excluded for producing a nonsense sentence). This likely has the greatest impact for anyone learning words that appear only one time, at which point it’ll be better to utilizing frequency lists based on a combination of series/volumes one is reading.
Orange is a great series, and any high-frequency vocabulary you learn along the way will translate to so many other series (with high school being a common setting in manga).
