5D Library › Spiritual Quotient Quantification, Prior-Conditioned Generation and Retrieval-Augmented Cognitive Agents · 中文
Preface: One Person and an 82-Volume Library in the Same Model
Note
This is not psychological assessment, medical or business advice and says nothing about whether any reading comes true; please read this first. The book studies the mathematics met by a system that hands a language model both a personal profile and an entire library: whether a five-dimensional “spiritual quotient” scale gives more information than one total, how much uncertainty an 8 KB personal profile can at most remove from the model’s answers, how a context window should be shared among profile, retrieved passages and conversation history, why retrieval for synthesis across books is hard, and whether the 82 volumes of this library are joined into one by their keywords. Every loading, correlation, reliability and retrieval parameter in it is a model parameter, not data from any assessment or user; the library graph comes from the public English keywords in each volume’s book.yaml.
This is the volume of 5D Library that puts one person and the whole library into the same model; its library analysis covers all 82 volumes, 001 to 082. The author’s outline makes three claims: first, that the spiritual quotient decomposes into five orthogonal dimensions (physiology and metabolism, neural emotion, game contracts, cycle risk, execution and growth, one for each of five “engines”), and that five dimensions explain more variance and carry more information than one; second, that injecting a personal profile vector of about 8 KB into the prompt together with retrieved passages lowers the conditional entropy of the answer, and that an optimal division of the context among profile, retrieval and history exists; third, that the knowledge graph of the library’s volumes has no islands. This book writes the three claims as runnable programs and tests them one by one.
Like the other books of 5D Library, it tries to refute first, then sets bounds, and reports negative results as they are.
What the book contains
Chapter 1 computes the eigenvalue spectrum of a five-dimensional scale and compares the explained variance and mutual information of five scores, one total and a single dimension. Chapter 2 computes reliability, the reliability of a profile’s “shape” and a blind-pick test. Chapter 3 computes exactly how much a profile lowers the conditional entropy of answers, and how that gain vanishes when the profile is wrong. Chapter 4 finds the optimal division of a context window with a Lagrange multiplier, and the optimal number of passages to retrieve. Chapter 5 simulates retrieval recall in an 82-volume library and questions that need three volumes at once. Chapter 6 builds a graph from the keywords of all 82 volumes and computes its Fiedler value, diameter and clustering, comparing them with random graphs. Chapter 7 reads the Zhouyi Xici, “looking up to observe the patterns of heaven, looking down to examine the order of the earth”, and Shao Yong’s method of observing things. Chapter 8 gives a seven-flag audit and a pre-registered blind comparison.
Claims corrected on checking
- Five dimensions are not always better than one; it depends on how they correlate. At a dimension correlation of 0.9, only 1 eigenvalue of the 20 items exceeds 1, and five dimensions are almost one (p01). The advantage of five scores is that they are “never worse”: when the outcome depends on all five, a single dimension explains only 0.0794 while the total and the five scores both explain 0.3968; when it depends on one, the total explains only 0.0794 and the five scores 0.3968 (p02). No gain exceeds half a bit.
- The uncertainty a profile can remove is bounded by the answer’s real dependence on it, not by the size of the file. In the book’s model, 243 profile states carry 7.9248 bits and an 8 KB file 65,536 bits, but the profile lowers the conditional entropy of the answer by only 0.8519 bits; when the answer does not depend on the profile, by 0 (p05). A profile wrong on a quarter of its dimensions keeps only 0.5350 of the gain (p06).
- The profile’s share of the context shrinks as the window grows. From 32k the profile sits at its 8 KB cap and everything else goes to retrieval and history (p07).
- Synthesis across books is a planning problem, not a retrieval problem. For questions that need one passage from each of three volumes, direct retrieval succeeds 0.073 of the time within the top 30; splitting the question into its three parts first succeeds 1.000 of the time (p10).
- “No islands” holds only after correction. In the 73-volume edition, volumes 3, 9, 24 and 31 shared no keyword with any other volume. Once they carry keywords their own text supports and volumes 080 to 082 are included, all 82 volumes form one component with a Fiedler value of 0.5475; without the keyword “pre-registration” the Fiedler value falls to 0.4161 and volumes 3 and 9 hang on 3 edges each (p11).
How to read the boxes
The boxes separate published evidence, runnable Python calculations (16 programs, standard library only, fixed random seeds, each run at build time with its real output printed beneath it), classical sources, notes and the author’s hypotheses; every hypothesis states what result would show it wrong. So that code stays intact in print, programs and their output use ASCII characters only; dimensions and schemes carry letters or English abbreviations, listed below.
Note
Code key: C, physiology and metabolism (engine I); D, neural emotion (engine II); G, game contracts (engine III); F, cycle risk (engine IV); E, execution and growth (engine V); SPREAD, the outcome depends on all five dimensions; ONE, on one only; NONE, answers ignore the profile; P / R / H, profile / retrieved passages / conversation history; PLAIN, direct retrieval; QUOTA, one passage per volume; SPLIT, retrieval after splitting the question.
Disclosure
The author uses the works of 5D Library, including this book, as the methodological basis of his community courses, the 5D Ascension Cards (Dimensional Shift Card Decks) and the Yafu AI knowledge base, and develops and runs paid online reading software and AI products on traditional divinatory and mind-body themes (a five-dimensional spiritual-quotient assessment, personal profiles, and retrieval-augmented question answering over this library); the models of this book apply directly to those products, so he has a direct commercial interest in its subject. The book promotes no software, course, card deck or service, uses no user or assessment data from the author’s businesses, and accepts no sponsorship. All its models are offered for mathematical study and teaching only and constitute no promise that any reading comes true and no psychological, medical, investment or other business advice. The experiment in Chapter 8 must be carried out with pre-registered code by an independent analyst who has no financial relationship with the author, and its results will be published whether positive or negative.
Fan, Y. (2026). Spiritual Quotient Quantification, Prior-Conditioned Generation and Retrieval-Augmented Cognitive Agents: Multidimensional State Vectors, Personalized Prompt Conditioning and Cross-Corpus Synthesis [五维灵商量化架构、先验条件化生成与检索增强认知体] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090203 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).