Case Retrieval

Chinese source (authoritative): ../../skills/case-retrieval/SKILL.md

Case Retrieval (Finding Similar Cases and Related Judgments)

Overview Table

Legal Disclaimer

Important notice:

1. The case-retrieval methodology and structured workflow provided by this skill are for reference only and do not constitute legal advice.
2. Mainland China’s legal system is a statute-based (成文法) system; cases generally lack universal binding force (except Guiding Cases (指导性案例)), and retrieval results should be treated as reference material rather than direct legal authority.
3. Guiding Cases issued by the Supreme People’s Court (最高人民法院) have “shall be referred to” (应当参照) effect and rank above ordinary cases.
4. When an AI system performs case retrieval, it must clearly mark case sources, hierarchical authority, and currency, and avoid citing adjudicative views that have been overturned or are no longer applicable.
5. Case-retrieval results should be reviewed by a practicing lawyer or other legal professional before use in actual legal matters.

I. Retrieval Foundations and Mindset

1.1 Description of Like-Case Retrieval (类案检索)

Like-case retrieval means searching for and applying cases that are similar to the pending case in basic facts, disputed issues, and legal application, and that have already taken effect as judgments of a people’s court. Under China’s statute-based system, like-case retrieval reflects integrated statutory and case-law thinking—using cases to help interpret abstract legal provisions and strengthen the persuasiveness of subsuming the minor premise under the major premise.

1.2 Three Retrieval Scenarios

By content, like-case retrieval roughly falls into three types:

1.3 Purposes of Retrieval

• Support an argument: Find adjudicative precedents supporting one’s own position
• Anticipate risk: Understand adjudicative trends and win rates in like cases
• Rebut the opponent: Find adjudicative views that negate the opponent’s arguments
• Fill gaps: Find judicial practice attitudes on issues not clearly provided by statute
• Unify understanding: Map the evolution of adjudicative rules on an issue
• Formulate strategy: Determine optimal litigation strategy through case analysis
• Draft documents: Provide case support for legal opinions / advocacy briefs

II. Pre-Retrieval Preparation: Problem Characterization

2.1 Background Information Collection

After preliminary analysis of the case evidence, distill the corresponding retrieval need. That need is, in substance, the legal or evidentiary issue the case must resolve.

Example: In a construction project dispute, besides claiming against the subcontractor who subcontracted the work to the actual constructor and against the developer as provided by judicial interpretation, may the actual constructor also claim project payment from others—e.g., a contractor who has neither a contractual relationship with the actual constructor nor a statutory obligation?

The retrieval need can be distilled as: “Must a contractor, by analogy to the developer’s position, pierce privity of contract and bear joint and several liability to the actual constructor? If so, must that liability, by analogy to developer rules, be limited to the unpaid project price?”

2.2 Information-Gathering Channels

When facing unfamiliar case types, drawing on prior research is the most efficient method:

1. Search-engine retrieval: Google, Baidu, etc., with keywords such as “XX adjudicative view, XX adjudicative holding, XX big-data report, XX compilation, XX cases, XX practical tips”
2. WeChat article search: Almost every specialized legal issue has been analyzed and compiled in WeChat public accounts
3. Preliminary database search: “Like-case retrieval” features on platforms such as PKULaw (北大法宝), Wolters Kluwer China Law (威科先行), and Faxin (法信)

Example: When handling a case, first distill from the evidence that the retrieval need is “determining authenticity of a company seal (公章).” Then search WeChat for “adjudicative rules on seal authenticity.” From cases cited in articles, extract the court’s terminology in judgments on seal authenticity—greatly helping keyword extraction for deeper retrieval.

2.3 Defining the Retrieval Scope

Case scope: Mainly typical cases published by higher courts and by the court concerned, and effective judgments rendered by them.

Time scope: Different time windows reflect current economic and social conditions and avoid results that diverge from present like-case reality. Except for Guiding Cases, prioritize cases from roughly the past three years.

Geographic scope: Besides typical cases and effective judgments of higher courts and of the court concerned, representative typical cases and effective judgments from provincial high courts and intermediate courts in other provinces/municipalities over the past three years may also be included.

III. Retrieval Elements and Keyword Extraction

3.1 Retrieval-Element Extraction Checklist

A. Basic elements (must extract)

• Cause of action / dispute type
• Core legal relationship
• Main disputed issues
• Applicable legal provisions

B. Factual elements (extract as far as possible)

• Party types (natural person / legal person / other organization)
• Nature and pattern of conduct
• Type and degree of harm
• Causal features
• Form of fault

C. Limiting elements (extract as needed)

• Geographic scope (nationwide / specific province / specific court)
• Time scope (recent years / specific period)
• Instance requirements (SPC / high court / intermediate / basic-level)
• Outcome tendency (uphold / dismiss)
• Amount-in-controversy range

3.2 Classification and Extraction of Keywords

Case-retrieval keywords roughly divide into “normative keywords” and “non-normative keywords”:

Four methods of keyword extraction:

1. Split from statutory text: For Company Law Article 63 (one-person company), extract “one-person limited liability company,” “shareholder,” “independent property,” “joint and several liability.”
2. Extract from case materials: Extract industry terms from contracts and documents (e.g., “illegal subcontracting,” “delayed completion”).
3. Distill from case facts: Distill keywords from factual fragments (e.g., “nominee shareholding,” “completed dry-share bribery”).

3.3 Keyword Combination Examples

Using “piercing the corporate veil among affiliated companies” as an example:

Group 1 (core concepts):
- "denial of corporate personality" + "affiliated companies"
- "piercing the corporate veil" + "personality confusion"
- "Company Law Article 20" + "joint and several liability"

Group 2 (types of confusion):
- "personnel confusion" + "affiliated companies"
- "financial confusion" + "independence"
- "business confusion" + "corporate personality"
- "three confusions" + "denial of corporate personality"

Group 3 (horizontal piercing):
- "horizontal piercing" OR "horizontal denial"
- "affiliated companies" + "joint and several liability" + "creditor"
- "actual controller" + "multiple companies" + "confusion"

Group 4 (source retrieval):
- "Nine Minutes" + "denial of corporate personality"
- "Minutes of the National Court Work Conference on Civil and Commercial Trials" + "Article 10"

IV. Retrieval Methods and Pathways

4.0 Data Sources and Tool-Calling Convention (Mandatory)

Core principle: Every case produced by this skill must come from real retrieval; generating cases from memory is strictly forbidden.

Before executing the pathways below, determine whether the runtime environment has case-retrieval tools (e.g., a connected case-library MCP service, retrieval API, or local case library):

1. If present: The designed search terms / queries must be submitted to that tool; its returned real results (case number, court, year, cause of action, adjudicative holding, source link) are the sole data source for subsequent similarity assessment and the report. Do not substitute or supplement results with “cases” from memory.
2. If absent: Still complete query design and methodological steps, but mark every concrete case reference [to be retrieved], and state clearly in the report: “No case library connected; the following queries must be executed manually.” Never fabricate case numbers, parties, or adjudicative holdings.
3. Tool-agnostic: This convention depends only on the abstract capability “input search terms → return real cases,” and is not bound to any vendor. Any equivalent case-library tool may be used. Concrete integration (e.g., PKULaw MCP) is described in README.md in this skill’s directory.

4.1 Retrieval Pathways

V. Case Similarity Assessment

5.1 Definition of Like Cases and Identification Markers

Under Article 1 of the Supreme People’s Court’s Guiding Opinions on Unifying Legal Application and Strengthening Like-Case Retrieval (for Trial Implementation): “Like cases are cases that are similar to the pending case in basic facts, disputed issues, legal application, and other respects, and that have already taken effect as judgments of a people’s court.”

Accordingly, determining whether a case is a “like case” requires these three identification markers:

Note: The three markers should be applied comprehensively, not in isolation. High similarity on one marker strengthens like-case recognition, but a fundamental difference on any marker may preclude treating the case as a like case.

5.2 Similarity Assessment Dimensions and Weights

For cases that pass preliminary screening, conduct deep similarity assessment along five dimensions. Each measures how similar the target case is to the retrieved case, and thus the case’s reference value.

5.2.1 Dimension One: Consistency of Legal Relationship (Weight 30%)

Meaning: Overall consistency between the target case and the retrieved case in the nature of the legal relationship, parties’ legal status, and rights–obligations structure. This is the threshold for treating a case as a “like case.”

Judgment points:

• Cause-of-action consistency: Same or most similar cause of action (e.g., both “dispute over shareholder liability for harming company creditors’ interests”).
• Parties’ legal status: Whether each party’s role corresponds (e.g., both involve a “one-person company shareholder” sued for joint and several liability).
• Rights–obligations structure: Whether the direction of core rights and obligations is the same (e.g., both involve a creditor seeking joint and several liability of a shareholder for company debts).
• Underlying legal relationship: Whether the underlying relationship generating the core dispute is the same (e.g., both arise from a sales contract creating a goods-payment debt).

Assessment tip: This dimension is the first threshold for like cases. If the legal relationship differs fundamentally, even high similarity on other dimensions makes the case unsuitable as a core reference.

5.2.2 Dimension Two: Correspondence of Disputed Issues (Weight 25%)

Meaning: Whether the core legal questions disputed by the parties are the same. The disputed issue bridges facts and legal application and is a key marker of a “like case.”

Judgment points:

• Identity of the disputed question: Whether the core legal question is the same (e.g., both concern “whether an audit report submitted by a one-person company shareholder can prove property independence”).
• Similarity of legal-application disputes: Whether the norms and interpretive questions align (e.g., both involve burden allocation under Company Law Art. 63).
• Number and layers of issues: Whether there is a single core dispute, or multiple disputes with the main one aligned (different secondary issues do not matter).
• Need for adjudicative guidance: Whether both sides seek the same type of rule guidance (e.g., both need clarity on “what evidence proves property independence”).

Assessment tip: This dimension is the core hub of like-case judgment. Even if legal relationships align, different disputed issues make adjudicative rules hard to apply directly. Focus on comparing the court’s listed “disputed issues” with those of the target case.

5.2.3 Dimension Three: Similarity of Key Facts (Weight 25%)

Meaning: Whether key facts that decisively affect the outcome are similar. Not all facts need match—only “element facts” matching legal constitutive elements are compared.

Judgment points:

• Element facts: Whether fact elements required by the norm are present (e.g., both involve “shareholder submitted complete annual audit report + bank statements + applied for judicial audit”).
• Decisive circumstances: Whether core circumstances affecting the outcome align (e.g., both involve “unblemished audit report; creditor offered no concrete confusion leads”).
• Factual defects and exceptions: Whether facts weaken or strengthen proof (e.g., one shareholder refuses original vouchers; the other proactively provides them).
• Spatiotemporal background similarity: Whether time, place, and trading customs of the conduct are close.

How to identify key facts:

• From the “facts found by the court” section, extract facts the court treated as “materially affecting the outcome.”
• Map to the statute’s constitutive elements to identify element facts.
• Exclude detail facts irrelevant to legal application (e.g., exact signing time or place of a contract).

Assessment tip: This is the most elastic dimension and must be judged against the specific norm. Subtle differences can reverse outcomes—assess with special care. Use the “substitution method”: replace key facts in the retrieved case with those of the target case and ask whether the outcome would still hold.

5.2.4 Dimension Four: Referenceability of Adjudicative Rules (Weight 15%)

Meaning: Whether the holding and legal-application rule distilled from the retrieved case can provide direct, clear guidance for the target case. This determines “usability.”

Judgment points:

• Clarity of the holding: Whether the court clearly stated a rule (e.g., “under condition A, legal effect B shall be found”).
• Transplantability: Whether the rule can apply directly to similar situations in the target case (the rule should be general, not highly fact-bound).
• Generality: Whether the rule is confirmed by multiple cases or is a singleton.
• Value as supplemental interpretation: Whether it concretizes an ambiguous provision (rather than merely repeating the statute).

How to extract adjudicative rules:

• Focus on rule-like statements in the “court’s opinion” section; hallmark phrases include: “shall be found to be…,” “shall not be upheld…,” “may refer to…,” “this court finds that … does / does not constitute…”
• Replace case-specific facts with typed descriptions to extract generally applicable standards.
• Rule format: “Under [condition/scenario], the [act/claim] of [subject] [shall/may/shall not] be found to constitute [legal effect].”

Assessment tip: This dimension determines practical persuasive force. One thoroughly reasoned, clear-rule case outweighs many vague ones. Prefer cases that concretize statutory provisions.

5.2.5 Dimension Five: Procedural-Issue Linkage (Weight 5%)

Meaning: Linkage between the target case and the retrieved case in trial procedure, evidence rules, and litigation stage. Procedural differences may affect the premises for applying an adjudicative rule.

Judgment points:

• Procedural consistency: Same second-instance, retrial, or first-instance procedure (second-instance and retrial cases are usually better reasoned).
• Evidence-rule application: Same burden allocation, standard of proof, etc.
• Litigation stage: Same stage (e.g., both main actions, or both enforcement-objection suits).
• Procedural defects and exceptions: Whether procedural issues affect the judgment’s force.

Assessment tip: Weight is low; usually meaningful only when the first four dimensions are highly similar. But procedure (e.g., rules from retrial cases) can markedly strengthen or weaken reference value. Especially: first-instance judgments not yet effective cannot serve as like-case authority; retrial cases (“民再”) are usually the most rigorously reasoned.

5.3 Comprehensive Assessment and Grouping

After assessing all five dimensions, combine similarity with hierarchical authority and judgment date to grade cases overall.

Comprehensive assessment points:

1. Judge comprehensively; avoid absolutizing one dimension: Combine all five; do not treat a case as a like case solely because one dimension is highly similar, nor exclude it solely because one dimension differs.
2. Sensitivity to key-fact differences: On some legal issues, a slight factual difference can reverse the outcome—pay special attention.
3. Equal assessment of adverse cases: Opposite-outcome cases also need assessment on these dimensions to clarify differences for rebuttal or risk forecasting.
4. Cross-validation across multiple cases: Single-case assessment may be biased; cross-compare multiple cases to identify mainstream rules and exceptions.
5. Combine with hierarchical authority: After similarity assessment, factor in authority level (Guiding Cases, Gazette cases, etc.) to determine final reference value.

Case grouping standards (reference):

VI. Case Screening and Ranking

Case screening is the core step in producing a like-case retrieval report. Scientifically selecting the most valuable cases strengthens persuasiveness and avoids overload. This chapter systematically sets out six screening standards (hierarchy, geography, time, procedure, background, importance) and three verification rules (effectiveness verification, risk verification), and explains judgment methods and application rules for each.

6.1 Case Screening Standards

6.1.1 Hierarchy Standard (Authority First)

Meaning: Authority level determines binding and persuasive force. The higher the level, the stronger the court’s duty to refer.

Judgment rules:

Practice tips:

• When earlier-tier cases suffice, generally do not submit later-tier cases.
• If earlier-tier cases are insufficient, backfill in order, but explain in the report.
• The report’s core goal is to “persuade by reason”; even only higher-court or same-court cases remain valuable if thoroughly reasoned.

6.1.2 Geography Standard (Jurisdictional Nexus)

Meaning: Territorial nexus between the adjudicating court and the court seized of the target case affects acceptability of the rule.

Judgment rules:

• Local court cases first: Cases from the same province, city, or even same court best reflect local adjudicative habits and judicial policy.
• Economically/culturally similar regions next: If local cases are insufficient, prefer courts in provinces with similar economic structure, culture, and judicial tradition (e.g., Yangtze River Delta, Pearl River Delta, Chengdu–Chongqing).
• Other regions last: General reference only.

Special notes:

• If the dispute type or key facts have clear regional features (local regulations, local policies, industry customs), prioritize cases from courts in the same jurisdiction; out-of-province value drops sharply.
• Example: rural land-contracting disputes—local regulations differ widely by province; prefer cases from the provincial high court or intermediate courts of the same province.

6.1.3 Time Standard (Currency)

Meaning: How close the judgment date and the time when facts arose are to the target case affects currency and applicability of the rule.

Judgment rules:

• Closer judgment dates preferred: Statutes, judicial interpretations, and judicial policy continually update; cases from the past 3–5 years best reflect current scales.
• Also consider when legal facts arose: Cases with close judgment dates but distant fact times may have applied different norms (e.g., Contract Law vs. Civil Code). Prefer cases whose fact times are close to the target case.
• Watch for legal change: If the target case should apply the Civil Code, but the retrieved case was decided in 2019 (under the Contract Law), confirm whether the Civil Code changed the rule. If unchanged, still usable; if changed, do not rely on it.

Currency score-reduction reference (for composite scoring):

• Cases 3–5 years old: modest reduction
• Cases 5–10 years old: larger reduction
• Cases over 10 years: generally do not use unless confirming the law is unchanged

6.1.4 Procedure Standard (Instance Rigor)

Meaning: Different trial procedures differ in rigor of reasoning and stability of views.

Judgment rules:

• “民再” > “民终” > “民初” > “民申”
• 民再 (retrial): Discussed by the Adjudication Committee or retrial amendment—most rigorous reasoning; highest reference value.
• 民终 (second instance): Final judgment; stable force; main reference source.
• 民初 (first instance): Judgments not yet effective (e.g., appeal still possible) do not qualify; effective unappealed first-instance judgments have lower value than second instance.
• 民申 (retrial review): SPC “民申” cases usually only reflect the original high court’s view without substantive SPC adjudication—lower value; generally not primary references.

Practice tips:

• When retrieving a first-instance instrument, always verify whether second-instance proceedings and outcomes exist.
• Prefer cases with “民再” and “民终” case numbers.

6.1.5 Background Standard (Special Endorsement)

Meaning: Trial background (panel composition, Adjudication Committee discussion) may confer extra authority.

Judgment rules:

• Panel members with legislative background: If panel members helped draft related judicial interpretations or judicial documents, their reasoning often tracks legislative intent more closely and has higher value.
• Decided after Adjudication Committee discussion: Language such as “decided after discussion by this court’s Adjudication Committee” signals professional endorsement and markedly strengthens the rule’s authority.

Practice tips:

• When reading judgments, note whether the caption or “court’s opinion” contains such language.
• If present, specially mark it in the report to strengthen persuasiveness.

6.1.6 Importance Standard (Gravity / Difficulty / Complexity)

Meaning: The case’s gravity, difficulty, and complexity affect typicality of the rule and depth of reasoning. Cases falling within situations that “shall undergo like-case retrieval” usually have higher value.

Judgment basis (referencing the SPC Guiding Opinions on Unifying Legal Application and Strengthening Like-Case Retrieval (for Trial Implementation) and related rules):

For the following four categories of cases, judges must conduct like-case retrieval; their judgments are usually more carefully considered and better reasoned:

1. Major, difficult, complex, or sensitive: Involving major interests, complex legal relationships, high social attention, strong policy character, etc. Such judgments often go through professional judges’ meetings or the Adjudication Committee; rules are more typical.
2. Group disputes or matters of wide social concern that may affect social stability: E.g., involving interests of many workers, owners, or investors. Such judgments often reflect judicial-policy orientation and have important reference value.
3. Possible conflict with like-case judgments of the same court or a higher court: Indicates adjudicative divergence; retrieved cases that unify the rule have higher value.
4. Reports that a judge engaged in unlawful adjudication: Such cases undergo supervisory review; fairness and reasoning are usually better assured.

Additionally, the following are mandatory or strongly recommended retrieval situations, and related cases likewise have high importance:

• Cases proposed for professional judges’ meetings or Adjudication Committee discussion: Rules usually emerge from collective discussion—higher authority.
• Lack of clear unified rules or no unified rules yet formed: Reasoning is often more detailed, seeking to fill gaps or clarify ambiguity.
• Cases where court/division leadership requires retrieval under trial-supervision authority: Reflects internal emphasis on case quality.

Practice application:

• When screening, prefer cases falling within the above. How to tell? Look for features such as:
  • “decided after discussion by this court’s Adjudication Committee”;
  • novel/difficult causes of action (e.g., “first case,” “first nationwide”);
  • many parties or public-interest involvement;
  • lengthy judgments with detailed reasoning.
• In the report, mark such cases (e.g., “This case was decided after Adjudication Committee discussion and has high reference value”).

6.1.7 Effectiveness Verification and Risk Verification

Before including any case in the retrieval report, complete the following three verifications. Cases failing any verification should generally not be listed (except opposite-outcome cases, which must be specially explained).

6.8.1 Confirm the Judgment Has Taken Effect

Method:

• Verify whether second-instance proceedings exist. If a first-instance judgment is retrieved, search China Judgments Online or related-case search by party name for a second-instance instrument.
• If second instance exists, confirm whether it conflicts with first instance (e.g., amendment or remand). If second instance affirms or only fine-tunes, the first-instance judgment may still be referenced, but preferring the second-instance judgment is better.

Risk notice:

• Judgments not yet effective do not qualify as references. First-instance judgments still within the appeal period or on appeal and not concluded cannot serve as like-case authority.

6.8.2 Confirm the Judgment Has Not Been Vacated

Method:

• Using main parties’ names as keywords (limit to core parties; avoid adding third parties and other non-core subjects lest results be missed), search related cases on China Judgments Online or databases.
• Check for retrial judgments, transfer-for-trial rulings, or third-party revocation judgments that vacated or amended the original judgment.

Risk notice:

• Risk of vacation via retrial or third-party revocation suit. If vacated, the adjudicative rule no longer has reference value.

6.8.3 Confirm the Judgment Poses No Risk

Method:

• Comprehensively review all content unrelated to one’s proof purpose, including:
  • Whether “facts found by the court” contains adverse factual descriptions;
  • Whether “court’s opinion” contains negative evaluations of similar claims;
  • Whether the outcome contains adverse findings (e.g., “the plaintiff was also at fault”).

Risk notice:

• Avoid adverse views in the case text. If a case overall supports one’s claim but contains an adverse detail (e.g., “although the shareholder was excused here, note that the shareholder and company must have no fund transfers”), opposing counsel may excerpt that part. Identify early and prepare a response, or consider replacing the case.

6.2 Case Ranking and Conflict-Handling Rules

When multiple like cases exist or conflict, rank and handle them by the “proximity principle.” The principle has three layers: proximity of level first > proximity of specialty first > proximity of time first.

• Proximity of level first: For the adjudicator of the seized court, cases from the immediate superior court have the greatest influence and persuasiveness because they directly relate to adjudicative risk (amendment or remand).
• Proximity of specialty first: Within the same court, cases from the specialized trial division hearing this case (e.g., finance, IP) have more value than those from other divisions of the same court.
• Proximity of time first: When conflict arises, the later-decided case prevails, because later cases usually reflect new statutes, judicial interpretations, or policy adjustments.

6.2.1 Direct-Line, Collateral, and External Cases (Foundation of Level Proximity)

…(truncated)