I have enumerated 6 rules.
(1) For each long article, divide the article into a maximum of a set number of major segments (about 10), and summarize each segment using a maximum of a set number of words (about 20).
I didn’t think it would come out to be this way - but I also realize this would end up with a word cap of 200 words to describe the entire text block - and it is actually a good idea.
(2) Then, one can choose to zoom in on either one or more of the divisions, and continue the splits.
(3) Of course, the preliminary step is for you to have someone who has read the whole text, and also, a trusted LLM model could do that part - provided that in the end, one would do their own analysis.
(4) I guess when making a text, such an index should be created recursively for each segment containing 200 or 300 words - whichever seems more appropriate (I haven’t decided).
For example: In a 500 page book with 200 words per page, first, one would make an index page, where the book will be divided into 10 segments. I initially thought to let the index page use the conventional layout with as many chapters as the author wishes, but then I realized that it would make the contents page too long and lead to many people ignoring it. Next, each segment - which I’d consider to be 50 pages each - to be divided into 10 segments of 5 pages each. Then, each such segment will also have an index describing each of the 5 pages. In this way, each 200 word segment gets a short description.
Now note that this division appeals only to instructive materials, and not to story books, for example.
(5) Now note that 10 is the maximum number of index segments. But a text could use a lower number of segments, since in the end, each 200 word segment will have an index too.
(6) As for content depth, after thinking about using various different systems, I’ve come to realize that content cannot always be made to be of a uniform depth, and therefore it should be allowed to grow without restrictions, and structure should only fit the content than the other way around.
So for example, if you consider the documentation for a game engine, it will have 10 main divisions, but division 7 may for example have a under division 8, about 7 divisions, and division 5 in it would have many divisions, going for up to about 7 layers of depth, while others would only have 2 or 3 layers of depth. In this case, the topic that goes deeper would be considered a vast topic compared to others.
(7) If in any case, a division of more than 10 elements becomes necessary, it should be considered an alternative index.
For example, Kashmir Shaiva Matham has 36 Thatthva-s, and even though one could group them into 3 divisions on one hand, or 5 divisions on another, they would still end up having more subdivisions, especially in the case of the classification into 3 divisions here, even when the direct list of 36 Thatthva-s would be easier to grasp. Or, maybe not, because when I first learned it, I started with the 23 Thatthva-s of Saankhya, just as Abhinavaguptha did too, and that was divided into the 3 Antah Karana-s, 5 Karma Indriya-s, 5 Jnaana Indriya-s, 5 Thanmaathra-s and 5 Maha Bhootha-s. So it used a classification into 5 groups for easy understanding too! And when it came to Kashmir Shaivism, Prakrithi and Purusha were added in, with the 23 and Prakrithi being Prakrithi Thatthva-s, then Purusha, the 5 Kanchuka-s and Maaya being the Maaya Thatthva-s or the Shakthi Thatthva-s, and the others being the Shiva Thatthva-s or the Eeshvara Thatthva-s. So I would say, yes, the classification into 10 components would be right.
The rationale for 10 is because a human has 10 fingers - and although some studies I had previously, roughly, seen had shown that an average human can only remember up to 7 things at a time, I would expect humans to do a bit more too - but then if the real answer was more than 7 - which is unlikely especially in this generation, I wouldn’t want people to more than 10. 10 is an okay spot, unless people disagree.