Task objective: Turn untested empirical judgments into theoretical expressions that have source levels, boundaries, and the ability to explain cases. First examine the proposition, then select the theory; first judge whether verification is necessary, then decide whether to use existing knowledge, local materials, or external sources.
Hard principles
- Examine the proposition first. Extract the original judgment, case, and usage scenario, and check for absolutization, causal leaps, definition substitution, overgeneralization, and reverse inference.
- Match quality takes priority. The theory must explain the core mechanism in the case. Fame cannot compensate for theoretical mismatch.
- Preserve the valid parts. When the case holds but the conclusion is excessive, narrow the boundary of the conclusion; do not overturn the whole passage just to switch theories.
- Person constraints are subordinate to task needs. When the user asks for "deceased famous people," only "confirmed deceased" and relevant professional identity are used as selection conditions; do not proactively add birth year, death year, or nationality changes. When existing knowledge is sufficient for a high-confidence judgment, it can be completed offline and the verification level marked; when there is controversy and it would affect eligibility, then treat external verification as a necessary step.
- Offline by default. First complete proposition review and theory matching based on the materials provided by the user, the current conversation, local materials, and the model's existing knowledge. Internet access is not a default research step, and one must not automatically go online just because one wants to supplement people, theories, original words, page numbers, or the number of candidates.
- Verify according to what the conclusion requires. Only enter external verification when external facts would materially change the conclusion, or when the user explicitly requests verification, sources, original words, page numbers, attribution of first authorship, or the latest information. When the user requests no internet access, internet access is prohibited; details that cannot be confirmed offline should be marked as "not yet verified," and user restrictions must not be violated on the grounds of verification.
- Distinguish classic formulations from verbatim quotations. When the user asks for "classic theory," "classic formulation," or "find a theoretical basis," it is by default permissible to deliver widely accepted concepts based on existing knowledge and clearly marked Chinese paraphrases; this does not equal requiring an internet search for verbatim original words. Only when the user explicitly requests "original words," "accurate source," "page number," or verbatim quotation should quotation verification be treated as a necessary part of the task.
- Layered backstage, Chinese only on the front end. During research, distinguish foreign-language original words, faithful translations, and Chinese paraphrases; when the original text has not been externally verified, prioritize clearly marked Chinese paraphrases, and do not package approximate wording from model memory as verbatim quotations. Documents given to course teachers, clients, or other parties do not display foreign-language original words.
- Give boundaries. Write at least one applicable scenario, one inapplicable scenario, and one counterexample that could overturn the current explanation.
- Refuse mechanical application. When no high-match theory can be found, directly state "no sufficiently matching theory has been found for now," retain the empirical judgment, or look instead for a school of thought or research conclusion.
- Fabrication is prohibited. When page numbers, original words, or first-authorship relationships cannot be verified, state the verification status and do not add speculative information.
- Distinguish the research process from the delivered document. Candidate scoring, document review, and search processes are for internal judgment. Versions shown to course instructors, clients, or other parties start by default from the "Overall Judgment," retain only a brief explanation of theory selection, and devote the space to the selected theories, case explanations, and revision suggestions.
- Control the number of theories. For the same set of cases, by default retain only the one main theory with the strongest explanatory power. Auxiliary theories must explain new mechanisms that the main theory cannot cover; if they yield only commonsense conclusions such as "express clearly" or "content must have value," delete them directly.
Execution Process
Step 1: Fix the original question
Output:
原命题：
原案例：
使用场景：
用户希望理论完成的任务：
For "task accomplished," select only one main item: explain the mechanism, increase credibility, revise the standard, organize cases, or generate PPT expression.
Step 2: Review the original proposition
Check item by item:
| Risk | Check question |
|---|---|
| Absolutization | Are words such as "all, necessarily, as long as, whenever, then" used? Can counterexamples easily hold? |
| Causal leap | Do the cases show only correlation, results, or feelings, yet are written as causes? |
| Definition substitution | Do the core terms used before and after change meaning? |
| Generalization from experience | Is a single case expanded into a rule for all scenarios? |
| Reverse inference | After seeing the result, is a single cause inferred backward? |
Make a status judgment:
保留: The original proposition matches the cases, and the boundaries are clear;收窄: The core insight is usable, but the wording or scope of application is too broad;重构: The cases are usable, but the original proposition cannot accurately explain them;放弃: Both the proposition and the cases lack support.
First write the revised standard, then proceed to the theory search.
Step 3: Define the mechanism the theory needs to explain
Rewrite "find a famous person" as 1-3 theory retrieval questions, for example:
- Why are concrete details more likely to form understanding or memory?
- Why are abstract conclusions effective in some scenarios and vague in others?
- What mechanism can explain both positive and negative examples?
The theory must answer the retrieval questions. Sharing only keywords with the case does not count as a match.
Step 4: Build a candidate pool
Propose 2-4 candidate theories and score them according to the table below:
| Dimension | Score | Judgment standard |
|---|---|---|
| Mechanism match | 0—3 | Can it explain the main contradiction in the case? |
| Reliable source | 0—2 | Is there an original work, paper, or authoritative archive? |
| Person recognizability | 0—2 | Is the target audience likely to understand the person's identity? |
| Usable for expression | 0—2 | Are there clear concepts, commonly used formulations, or Chinese paraphrases that can be accurately annotated? |
| Clear boundaries | 0—1 | Can it explain when it holds and when it fails? |
A total score below 7 cannot be selected.机制匹配 < 2 or 来源可靠 < 1 is eliminated directly. When the user requests a deceased person, "the death status can be confirmed with high confidence, and the current verification level is truthfully annotated" is an additional hard threshold.
After scoring is complete, perform deduplication once more:
- When two theories explain the same mechanism, keep only the one with the higher mechanism match score;
- An auxiliary theory must explicitly state "the part that the main theory cannot explain but it can explain";
- When a new theory can only generate broad, correct conclusions with no information gain, delete that theory;
- When the case can still be fully explained after deleting the theory, do not keep it just to increase the number of famous figures.
Candidate comparison is used for internal screening. When the user explicitly requests the research process, output the complete table:
| 候选人物 | 候选理论 | 能解释什么 | 解释不了什么 | 得分 | 结论 |
|---|---|---|---|---:|---|
Step 5: Determine figures, theories, and verification levels
First determine whether the current task truly requires external verification:
- When only explaining mechanisms, revising propositions, or finding classic theories is needed, use the user's materials, local sources, and the model's existing knowledge by default, and write the verification status as "based on existing knowledge, not externally verified in this round";
- When the user requests a "classic formulation" but does not request verbatim original words, prioritize outputting a Chinese paraphrase and clearly mark it as "paraphrase";
- When the user explicitly requests accurate original words, page numbers, editions, attribution of first authorship, or disputed facts, then decide whether to perform external verification according to the "Knowledge, Verification, and Internet Standards" at the end;
- When the user explicitly requests no internet access, delete unnecessary precise details and retain the theoretical original meaning, case explanations, and applicable boundaries that can be completed offline.
Figure positioning
In the main text for the client, retain only the country or main academic affiliation, relevant identity, Chinese name, and common foreign name. Birth and death years, career history, awards, nationality changes, and related contributions are used for background verification and do not enter figure positioning or the PPT.
Recommended sentence pattern:
[国家或主要学术归属][相关身份][中文名]（[外文通行名]）
For example:
美国语义学家塞缪尔·早川（S. I. Hayakawa）
Chinese names follow these rules:
- Prioritize the translated name used in the person's own Chinese signature, official Chinese publications, or authoritative sources such as the National Library / universities / publishers;
- When there is no unique authoritative translated name, use the common translated name in Chinese publications and write "commonly translated in Chinese as...";
- When two common translated names exist, use the most common one in the main text and write "also translated as..." in the notes;
- When the Chinese translated name cannot be verified, retain the foreign name and clearly write "no reliable unified Chinese translated name has been found for now"; transliteration on one's own is prohibited;
- When birthplace, ethnicity, nationality changes, and country of employment are easily confused, do not rely on unverified precise details to complete figure positioning; when necessary, mark as "not yet verified" and do not include it in the main text for the client.
- When the user requests "deceased figures," prioritize figures whose identity is clear and relatively uncontroversial in existing knowledge, and do not proactively add birth and death years. Only when the death status itself is uncertain and affects eligibility should external verification be treated as a necessary fact that may change the conclusion.
Theory card
理论名称：
理论要解决的问题：
核心机制：
原始出处：作者，《作品》，年份；只有确知时才补版本 / 章节 / 页码
核实状态：用户材料 / 本地资料 / 基于已有知识，未在本轮外部核实 / 原始来源已核实 / 权威二手来源已核实 / 尚未核实
During research, divide quoted text into three layers:
外文原话：只有实际查阅并核实时才记录，不进入给当事人的文档
中文译文：只有实际核对原文时使用，尽量保留原意，不添加案例结论
中文意译：可依据可靠的理论原义和已有知识压缩表达，必须明确标注「意译」
The theory card for the client outputs only the Chinese translation or Chinese paraphrase. When the original text has not actually been checked, it must not be marked "Chinese translation" and can only be marked "Chinese paraphrase." When page numbers vary by edition, also write the edition; when only the chapter can be confirmed, cite the chapter and do not guess the page number; page numbers that do not affect the core conclusion may be omitted directly.
Step 6: Reinterpret the case using theory
Map sentence by sentence; do not merely attach the theory name in front of the case.
| 案例原句 / 事实 | 理论中的对应概念 | 理论解释 | 修正后的判断 |
|---|---|---|---|
Check the mapping:
- Whether the theory explains why the case is effective;
- Whether the theory explains why the original statement overstated things;
- Whether the case truly provides the evidence required by the theory;
- Whether the new interpretation introduces facts that do not exist in the case.
Step 7: Write the applicable boundaries and counterexamples
适用场景：
- ...
不适用场景：
- ...
反例：
- 如果出现……，当前解释需要被收窄或放弃。
仍需核实：
- ...
No theory can lead to infinite conclusions such as "copy should always be more specific" or "all success is caused by a single reason."
Step 8: Generate the standard output
First determine the reader:
- For researchers or skill debuggers: candidate scores and the verification process may be attached;
- For course instructors, clients, or other parties involved: use the streamlined delivery version by default, and do not show file checks, decryption processes, or the complete elimination record.
Deliver the document for the party involved according to the following template:
# 1. 总体判断
用一小段或一张简表说明哪些命题保留、收窄、重构或暂不补理论。
# 2. 理论选择说明
用 1 段说明入选标准与主要淘汰原因，不展开评分过程。
# 3. 理论锚点
原命题：
审查状态：保留 / 收窄 / 重构 / 放弃
修正后的标准：
## 人物定位
用一行写「国家或学术归属＋相关身份＋中文名＋外文名」。
## 相关理论
理论名称：
理论原义：
中文译文：
原始出处：
核实状态：
## 理论如何解释案例
| 案例 | 理论解释 | 得出的判断 |
|---|---|---|
## 适用边界
适合：
不适合：
反例与待核实事实：
## PPT 可用版
【标题】10 字以内
【人物署名】国家或学术归属＋相关身份＋中文名（外文通行名）
【理论表述】已核实的中文译文，或基于已有知识且明确标注的「意译」
【一句解释】20 字以内
【案例对比】两侧各不超过 30 字
【演讲者注】说明边界、出处和不能上屏的核实信息
When the same document has multiple theoretical anchors, "Overall Judgment" and "Theoretical Selection Explanation" appear only once, followed directly by each theoretical anchor. The core theory is placed first, and auxiliary theories follow.
Knowledge, Verification, and Internet Standards
Do not use the internet by default. First answer based on the materials provided by the user, the current conversation, local materials, and the model's existing knowledge. Theory matching, proposition review, case reinterpretation, and boundary judgment can usually be completed offline; do not use the internet to make the answer appear more academic, fill in theory card fields, or increase the number of candidates.
Internet use is permitted only when any of the following conditions is met:
- The user explicitly requests an internet search, fact verification, accurate sources, verbatim quotations, page numbers, editions, attribution of first invention, or the latest research;
- A certain external fact would materially change whether the main theory is selected, whether a person meets hard constraints, or whether the core conclusion holds;
- The user requests precise quotations suitable for formal publication, and the existing materials and local materials are insufficient for verification.
The following situations cannot trigger internet use on their own:
- The user only says "find classic theories," "find classic expressions," or "find support for an opinion";
- Existing knowledge is sufficient to complete mechanism matching, and only page numbers, verbatim wording, or non-critical biographical details are missing;
- The goal is to confirm more candidates, supplement more cases, or improve the surface completeness of the answer;
- It is already possible to accurately express the original meaning of the theory through Chinese paraphrase.
When the user explicitly requests no internet use, it is prohibited to call web search, browsers, network APIs,curl or any other external retrieval capability. In this case, you should:
- Normally complete proposition review, theory matching, case explanation, and boundary analysis;
- Write the source status as "based on existing knowledge, not externally verified in this round";
- Delete precise page numbers, editions, birth and death years, and verbatim quotations that cannot be confirmed and do not affect the conclusion;
- Mark facts that would affect the conclusion but cannot be confirmed offline as "not yet verified," and do not fill them in with guesses.
When internet use is needed, first explain in one sentence "what key facts need to be verified, and why they would change the conclusion," then search according to the principle of minimum sufficiency. Stop immediately once the evidence is sufficient to support the conclusion; if the same fact cannot be verified twice in a row, mark it as "not yet verified" and stop.
When conducting external verification, the source priority is:
- Original works, original papers, author archives;
- Universities, academic societies, governments, museums, authoritative publishers;
- Peer-reviewed papers or reliable academic reference works;
- High-quality secondary explanations;
- Encyclopedias are only used to discover leads, and quotation sites and aggregator articles cannot be used as the sole source.
Only when the task actually enters external verification, seek sources separately as needed for the following two types of key facts; when it has not entered external verification, do not proactively search just to fill in the format:
- Personal identity, birth and death dates;
- The original meaning of a theory, the original words, or the source of a work.
If a quotation is found through secondary literature, write "this secondary literature quotes from..." and do not disguise it as having directly consulted the original work.
Calibration example
Input:
「需要解释的词就不是好文案。」
「我努力创业，才有了今天的成就。」
对比案例是「我开了 30㎡ 的重庆小面馆，2 个月做到 35 万」。
请找一个已故名人的理论解释。
Minimum acceptable result:
- By default, do not connect to the internet; first use existing knowledge to complete the theory match, and clearly write the verification status;
- The original proposition is judged as
收窄, because abstract expression can serve as a title, summary, and value judgment; - The retrieval question focuses on "how abstraction and concreteness divide the work";
- Hayakawa's "ladder of abstraction" takes priority over theories that only explain memory effects;
- Write the person positioning as "American semanticist Samuel Ichiye Hayakawa (S. I. Hayakawa)" and do not output a field-style person profile card, birth and death years, or detailed resume;
- Output the person's Chinese translated name, field positioning, Chinese paraphrase of the theory, known source, and verification status; use "Chinese translation" only after actually checking the original text;
- Explain that specific cases improve perceptibility, but "350,000" still needs its basis verified;
- The Chinese paraphrase must be clearly marked and must not be passed off as a faithful translation of Hayakawa's original words.
Self-check
- Is the proposition examined first, and then the person sought?
- Does the selected theory pass the scoring hard threshold?
- For the same question, is only the main theory with the strongest explanatory power retained?
- Does the auxiliary theory provide a new mechanism not covered by the main theory?
- By default, is the answer completed first using user materials, local materials, and existing knowledge, without treating internet access as the starting point of theoretical research?
- When the user only asks for a "classic formulation," is it avoided to automatically interpret this as verbatim quotation verification?
- When the user requests no internet access, are web pages, browsers, network APIs, and command-line network access completely stopped?
- If the person's birth and death dates, country, and identity have not been externally verified, are non-key details omitted or the verification status truthfully marked?
- Is the Chinese translated name marked with the current verification level, and truthfully marked when there is no unified translated name?
- Does the person positioning retain only country or academic affiliation, field, Chinese name, and foreign-language name?
- Is a distinction made in the backend between original words, translation, and paraphrase, and are foreign-language original words hidden in the document given to the client?
- Is it avoided to mark secondary quotations as direct quotations from the original work?
- Are cases mapped sentence by sentence, rather than only attaching theory labels?
- Are the applicable boundaries, counterexamples, and facts pending verification written out?
- Does the document given to the client omit file checks, the decryption process, and complete candidate scoring?
- Is the PPT copy short, and does it retain source information in the speaker notes?
After completing the current task, end directly. Only when the user explicitly asks about the next step and the current environment already has /dbs installed, briefly prompt to enter /dbs。