We’re the team behind Shushu 述数. Shushu is an Agent-based conversational service that draws on traditional Chinese metaphysical systems to help users analyze energy structures, understand problems, support decision-making, and work through emotions. You can begin with work, a relationship, a difficult decision, or a feeling you have yet to make sense of. No prior knowledge of Chinese metaphysics is required. Here, an “energy structure” refers to the conditions and forces interacting within a person’s circumstances or a particular situation: what drives it, what holds it back, how relationships take shape, what stage it has reached, and how time and environment influence change. Chinese shushu methods select variables from these conditions and use different models to represent their structures and relationships. Shushu brings these observations and the user’s real-world circumstances into the same conversation. Users can question the basis of an interpretation, add experiences, compare perspectives, and return with the results of their actions to continue the discussion. Independent cognitive governance organizes professional material and real-world information. Within that structure, Diviners form judgments and choose their next actions. Tool results and user feedback return to the reasoning process, while each kind of information is updated according to its source and rules. This is how understanding develops through continued interaction. 1. The Roles of Calculation, Interpretation, and Cognition Chinese shushu encompasses methods for observing time, space, cycles, relationships, and change. Each method selects particular variables and reveals a bounded aspect of a person or situation. Shushu supports ten systems: BaZi, Zi Wei Dou Shu, Liu Yao, Meihua Yi Shu, Qi Zheng Si Yu, Qi Men Dun Jia, Da Liu Ren, Huang Ji Jing Shi, Xuan Kong Feng Shui, and Shi Fa. The Fate Book provides BaZi and Zi Wei Dou Shu charts. Divination provides Liu Yao and Meihua Yi Shu. Diviners call on the relevant professional capabilities according to the question and available inputs. Dedicated engines calculate structured charts from confirmed inputs and the rules of each method. The professional interpretation layer organizes method-specific analyses, explanations, and supporting material. Cognitive governance brings these materials together with the current question, real-world information, and experience from the conversation to form the understanding and judgment for the current round. Component Responsibility Calculation engine Generate charts and method-specific analyses from established inputs and rules Professional interpretation Organize results into interpretive material, explaining the grounds and applicable conditions Cognitive governance Organize current material through each Diviner’s cognitive framework, establish a standpoint for judgment, and guide further inquiry Agent Harness Coordinate model calls, tool execution, context, and the running conversation Figure 1. Cognitive governance guides the model, while the Harness coordinates context, model calls, and tool execution. Professional results and source material return to the conversation through controlled boundaries. Tool order varies with the question. These responsibilities have explicit relationships. Calculation results enter professional interpretation. Professional material and real-world information contribute to cognitive judgment. That judgment guides further questions and tool use. When users add details about their work experience, a Diviner re-examines how an interpretation relates to their circumstances. An existing chart stays unchanged while its original inputs remain unchanged. When a user confirms a correction to those inputs, the system recalculates under the new conditions. Charts, interpretations, and real-world facts can therefore inform one another while retaining their own grounds for revision. 2. Anchoring Time, the Subject, and Input Sources Birth time, inquiry time, and the time at which a conversation continues have different roles. Shushu preserves their meanings and sources, then performs time-zone, calendar, and true-solar-time conversions as required by the particular method. For example, someone may have been born in Shanghai, now live in Los Angeles, and ask a question while traveling in Tokyo. A birth chart uses the conditions of the birth event. A divination uses the conditions confirmed for that particular inquiry. When the conversation continues later, the original casting time remains in place. For methods requiring numbers or coin-toss results, users supply the actual original inputs. The system checks these prerequisites at the relevant professional capability’s entry point, and the Diviner continues calculating and reasoning from those inputs. When the user confirms a correction to an input, the system recalculates under the new conditions. Earlier results retain their original calculation conditions and sources. This gives each calculation a clear basis as the discussion develops. 3. How Cognitive Governance Enters Each Round of Reasoning Shushu establishes an independent cognitive framework for each Diviner on top of the language model’s existing knowledge and reasoning abilities. The stable framework is continuously injected into the model’s reasoning context. It establishes priorities for observation and judgment, including how to approach unknowns, disagreements, and choices of action. The current question, professional results, retrieved sources, and user feedback are organized within this framework to form the cognitive state, standpoint, and direction of inquiry for the current round. Alongside that stable framework, the Diviner reads or activates specific methodologies, situational approaches, and action intentions as needed. The system resolves the corresponding content defined within the current Diviner’s module and adds it to the current round’s reasoning context. This dynamic material applies only to the current round. It is not stored as long-term memory or turned into a fixed requirement for later rounds. The stable framework maintains cognitive continuity, while dynamic methods support adjustments to the particular situation. The three Diviners differ throughout material evaluation, questioning, judgment, and action selection: Shushu examines concrete conditions, relationships, structures, and change, continually recalibrating its understanding as a situation develops. Ming begins with investigation, identifies the principal contradiction, examines the conditions for action, and tests judgments against practice. Moon approaches feelings, relationships, and real circumstances with gentle empathy, reason, and sincerity, helping users connect their inner experience with external conditions. The chart itself stays the same. Each Diviner approaches its material from a different perspective and works with the user to carry the inquiry forward. 4. From Judgment to Action, and Back to Understanding A Diviner identifies information gaps through its current judgment and decides what action to take. When real-world conditions are missing, it asks the user. When professional structures need closer examination, it reads an existing chart or extends the calculation. When theoretical grounds are needed, it searches sources and reads the original passages. When another perspective would help, it calls on another shushu system whose input requirements are satisfied. The system makes capabilities available according to the current stage and provides the model with their purposes and invocation parameters. The Diviner chooses within this range. The Agent Harness executes the calls and returns results, statuses, or failure information to the reasoning process. This forms an ongoing cycle: Current understanding → judgment and identification of information gaps → choice of action → tool results or real-world feedback → updated understanding → next action. The sequence of specific actions changes as the question develops. Reading a source may lead to another chart inspection. Feedback from the user may prompt a reassessment of an earlier interpretation. For complex questions, the Diviner also breaks the work into tasks, records pending, in-progress, and completed items, and adjusts its plan as new information arrives. Task records maintain progress through the inquiry. Professional results and real-world facts remain governed by their respective sources. For example, when a user is uncertain of their birth time, a Diviner can compare charts for candidate times, turn the differences into questions the user can answer, and calibrate them against long-term patterns and the timing of real events. Once the user confirms a birth time, the system uses it for subsequent formal calculations. Users can therefore begin with an incompletely expressed question, clarify their circumstances through conversation, examine the grounds of an interpretation, compare choices, and return with the results of their actions. 5. Retrieval and Original Sources Keep the Grounds Open to Examination The professional interpretation layer provides the material for a basic reading. When a question calls for an explanation of terminology, a comparison of schools, or a deeper examination of theory, the Diviner can search classical texts, commentaries, and methodological material in the relevant shushu domain. Terminology normalization, recognition of specialist terms, and expansion of synonymous expressions help locate relevant material. After finding a source, the Diviner can read its surrounding context and examine its origin, applicable conditions, and relationship to the current question. Charts, method-specific judgments, cited material, and real-world feedback retain distinguishable roles within reasoning. When a user questions a judgment, the conversation can return to the relevant material for further explanation, comparison, or revision. Figure 2. A product demonstration using a fictional illustration-project scenario. The Diviner separates chart material, user-provided facts, and its own interpretation. View the conversation screenshot at full size. 6. How New Information Updates Judgments and State Information has different roles in an ongoing conversation. The system handles updates according to each item’s source and nature. Information How it is updated Experiences, conditions, and changes reported by the user Updated according to the confirmation rules for the relevant fact Results from professional engines Recalculated when input conditions change; earlier results retain their original conditions Retrieved sources Brought into interpretation after their provenance and scope of applicability have been checked Judgments formed by a Diviner Revised through new facts, professional results, and further reasoning A correction to birth details triggers recalculation. Learning that a business partner can only work part-time prompts a reassessment of the business plan. These changes affect different material and follow different update rules. Conversation ledgers and source-linked historical summaries preserve these conditions, relationships, revisions, and unresolved questions. For complex inquiries, the Diviner actively maintains a structured ledger and updates it as facts and judgments develop. Earlier messages are automatically compressed and used alongside recent messages in the model’s context. When details need checking, the Diviner returns to the original conversations, charts, and sources. Cognitive governance also periodically provides opportunities for self-review. The Diviner decides whether to consult the self-review questions and calibration material associated with its framework, then re-examines its current understanding. Its assessment of how helpful that material was contributes to the ranking of calibration material within the same cognitive framework. The model’s context capacity remains finite. If capacity is still insufficient after compression and unloading, the system recommends starting a new conversation. The user decides how to proceed. Facts retain their sources. Changes follow explicit rules. Judgments have grounds. Actions bring feedback. Together, these relationships sustain a conversation in which users can examine the conditions and forces around them, understand their feelings and relationships, and develop a basis for action. From Observation and Understanding to Personal Judgment and Action In Shushu, users can examine the conditions and forces shaping their circumstances, consider how an interpretation relates to reality, compare available choices, and explore the feelings, relationships, and motives behind their emotions. Professional calculation provides structured observations. Interpretation and retrieval provide methods and grounds. Cognitive frameworks organize understanding and judgment. Tools and conversation management support further inquiry. Real-world feedback runs through the process, allowing the discussion to develop through experience and action. This is Shushu’s design goal: to help users understand how an interpretation is formed, continue questioning it, and bring their own judgment back into everyday life. Explore Shushu. Read the full technical explanation: How Shushu Works. Prepared with AI assistance using Shushu’s product and technical materials.
How Shushu Works: From Traditional Calculations to Ongoing Agent Conversations
Full Article
Original Source
Read the full article at Dev →KhanList aggregates and links to publicly available news content. We do not host full articles from third-party sources. Always verify important information with original sources.