Legal Rule Induction: Towards Generalizable Principle Discovery from Analogous Judicial Precedents

Baixuan Xu; Chunyang Li; Haoran Li; Tianshi Zheng; Wei Fan; Weiqi Wang; Weixing Shen; Yangqiu Song; Yiran Hu; Zheye Deng

REVIEW 4 major objections 6 minor 22 references

Legal Rule Induction: Towards Generalizable Principle Discovery from Analogous Judicial Precedents

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The authors present LRI, a task and benchmark for inducing legal rules from analogous Chinese judicial precedents, and show that fine-tuning improves LLM performance significantly.

desk verdict First real benchmark for legal rule induction, with a plausible but judge-dependent fine-tuning win; dataset not released. read the letter →

arxiv 2505.14104 v1 pith:T4ZNXIFM submitted 2025-05-20 cs.CL

Wei Fan , Tianshi Zheng , Yiran Hu , Zheye Deng , Weiqi Wang , Baixuan Xu , Chunyang Li , Haoran Li

show 2 more authors

Weixing Shen Yangqiu Song

This is my paper · ORCID

classification cs.CL

keywords legalrulescasesllmsprecedentsrulesetsanalogous

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper introduces a new task called Legal Rule Induction (LRI). Given six to ten court decisions that are analogous to each other, a computer program must produce a small set of general legal rules that summarize the shared preconditions, behaviors, and consequences. The authors build a training set automatically from over nine million Chinese court judgments published online: cases citing the same legal articles are grouped, and a large language model called DeepSeek-R1 is used to write candidate rules. After filtering, this yields 5,121 case sets for training and a separate expert-annotated test set of 216 case sets. The authors evaluate many large language models on this task. They find that models often produce rules that are too broad or contain made-up content. A proposed iterative pipeline, SILVER, improves recall. The most striking result is that small open-source models, after fine-tuning on the new training set, nearly match or beat much larger commercial models on the test set. The authors are careful to exclude statutory citations from the inputs so that models cannot simply recall the law, and they use a human-annotated gold standard for the test set. However, the scores used to measure quality come from an automated judge, DeepSeek-V3, whose assessments were checked by humans on only 114 rules. Since the training rules were also generated by DeepSeek models, the reported gains could partly reflect stylistic alignment with that judge.

Extended reading notes

Core claim

The paper's central claim is that LRI is a well-defined, benchmarkable task and that training on LRI-AUTO markedly enhances LLMs' rule induction capability: fine-tuned 3-8B models reach Mic-F1 70.9-73.2%, outperforming larger closed-source models under direct prompting (Section 5.2, Tables 2-3).

Load-bearing premise

The reliability of the DeepSeek-V3 judge: all reported F1 scores hinge on the judge's ability to determine logical equivalence between predicted and gold rules, and only 114 judge decisions were manually verified (Section C.3, Table 5). If this judge is biased toward DeepSeek-generated rule styles, the headline improvements from fine-tuning could be inflated.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Two things to know before you read this. First, it's the first serious benchmark for legal rule induction (LRI): case sets labeled with expert-annotated explicit and implicit rules, plus a large auto-curated training set. Second, all F1 numbers are mediated by a single LLM judge (DeepSeek-V3) whose equivalence decisions were spot-checked on only 114 rules, and the training rules were generated by DeepSeek-R1, so a same-family stylistic bias is a live concern.

What's genuinely new: the task formalization (three-element rule structure: condition, behavior, consequence), the LRI-AUTO and LRI-GOLD datasets (38k Chinese cases in 5,121 auto-annotated sets, 216 expert-gold test sets), and the SILVER iterative induction-verification pipeline. The curation is careful: statutory citations are removed from input, PII is anonymized, explicit rules must apply to all cases, implicit rules to a majority, and a human audit of 100 auto sets reports 93-99% quality on several dimensions. That is real work and real evidence.

The experiments are informative, too: direct LLMs over-generalize and hallucinate; CoT mostly trades recall for precision; SILVER boosts recall; and LoRA fine-tuning on LRI-AUTO lifts 3B-8B models from around 19% to 71-73% Micro-F1, beating several larger closed models in the direct setting. That is the headline result.

The soft spots are mostly in evaluation. The judge is one model, DeepSeek-V3, and the manual validation is only 114 decisions across 38 model/settings — about 3 per condition, and for the four LoRA models it's 12 total. That is not enough to rule out per-model bias, and since LRI-AUTO rules were generated by DeepSeek-R1, fine-tuned models inherit DeepSeek-style phrasing that the judge may treat as logically equivalent more often. I don't think this kills the paper — the qualitative finding that fine-tuning helps is consistent across four models and multiple metrics — but the precise F1 deltas should be treated as approximate until there is a second judge or a larger human re-evaluation.

Also: the dataset and code are not released, which is a major reproducibility gap for a benchmark paper. And the abstract's 'over 76% gains' doesn't match the tables (best LoRA Mac-F1 is 70.73). Minor, but sloppy. The Chinese-only scope is acknowledged in the limitations.

Who should read it: anyone working on computational law, legal NLP, or inductive reasoning in LLMs. It deserves a serious referee — I'd send it to review with a request for a second judge, a larger judge validation, and dataset release. Recommendation: engage with it, but push for those revisions before acceptance.

Referee Report

4 major / 6 minor

Summary. The paper formalizes a new task, Legal Rule Induction (LRI), in which a model must derive concise normative rules from a set of analogous judicial precedents, structured as hypothetical condition, behavior pattern, and legal consequence. The authors construct LRI-AUTO, a large automatically built dataset of 5,121 case sets (38,088 Chinese cases) with rules extracted by DeepSeek-R1, and LRI-GOLD, a smaller expert-annotated test set of 216 case sets. They evaluate direct prompting, chain-of-thought, long-chain-of-thought, and an iterative induction-verification pipeline (SILVER) across several LLMs and reasoning models, and fine-tune four 3B-8B models with LoRA on LRI-AUTO. The central empirical claim is that LoRA fine-tuning raises small models to Micro-F1 scores of 70.96-73.18%, surpassing larger closed models under direct prompting. Rule-level evaluation uses DeepSeek-V3 as an automated judge for logical equivalence, with manual validation on 114 judge decisions.

Significance. The task definition and benchmark are a useful contribution: LRI-GOLD is expert-annotated, the three-element rule schema is principled, and the authors include human quality audits of both the automatic dataset and the judge, plus detailed prompts. The paper also makes several careful design choices, such as removing statutory citations from inputs and filtering rules by applicability. If the headline results survive a stronger judge-validation protocol, the finding that small fine-tuned models outperform much larger models on this inductive task would be practically meaningful. The main risk is not the task definition but the reliability of the automated judge for the central comparison.

major comments (4)

[Section C.3, Tables 2 and 3] The claim that fine-tuned 3-8B models outperform larger models (Section 5.2) rests entirely on the DeepSeek-V3 judge, but the manual validation of the judge is too thin to rule out a family-specific bias. Only 114 judge decisions were manually checked in total (3 per model/setting), and for the four LoRA models this amounts to 12 audited decisions. Because LRI-AUTO rules were generated by DeepSeek-R1 and the judge is DeepSeek-V3, models trained on LRI-AUTO may produce rules in a DeepSeek stylistic register that the judge is systematically more likely to declare logically equivalent to gold rules, inflating the reported Mic-F1 gains. The aggregated audit in Table 5 cannot detect such per-model bias. I request a per-model audit of judge decisions (especially for the four LoRA models), a comparison with a second judge from a different model family (e.g., GPT-4o or Claude), and a report of the distribution of judge verdicts with an error analysis.
[Section 5.2, Tables 2 and 3] The headline comparisons are presented without any measure of variability. Each F1 number appears to come from a single evaluation pass, and the fine-tuned models differ from each other by only 2-3 F1 points (e.g., 70.96, 71.70, 73.18). Without repeated runs, bootstrap confidence intervals, or a significance test, the paper cannot support the 'markedly enhances' claim beyond the large pre/post gap, and even that gap could be affected by judge noise. Please report variance over at least a few independent runs or a bootstrap over case sets.
[Equations (1)-(2), Table 15] The rule-level micro-F1 treats each predicted rule as an independent unit, but the judge's 'logical equivalence' criterion is binary and the 'partial match' category is collapsed into incorrect. This makes the metric very strict and potentially sensitive to the rule set size. The paper should report the frequency of the judge's verdict categories (equivalent / partial / non-match) for at least a sample of models, and should state explicitly how partial matches are handled in the reported F1, since the judge prompt in Table 15 lists 'Partially Matches' but the scoring rule treats only 'Logically Equivalent' as correct.
[Section 4.3, Section 4.4] The LRI-AUTO filtering inherits errors from DeepSeek-R1's applicability assessments, since both explicit-rule 100% applicability and implicit-rule >50% applicability are judged by the same model that generated the rules. The human audit of 100 sets reports 93-95% field correctness, which is reassuring, but the audit is not tied to the filtering step and does not measure the impact of the residual noise on fine-tuning. Please provide a breakdown of the audit by rule type and report whether the cases where 'all fields are correct' fails are concentrated in implicit rules.

minor comments (6)

[Abstract] The phrase 'over 76% gains' is ambiguous; it should specify whether this is the absolute percentage-point improvement or the relative improvement, because the numbers in Table 3 (e.g., 19.21 to 70.96) correspond to much larger relative gains.
[Table 2 caption] There is a typo in the caption: 'Webold the best' should read 'We bold the best'.
[Figure 2] The labels 'TrainInfer' and 'Infer' in the bottom-right of Figure 2 are awkward; consider 'Training' and 'Inference' for clarity.
[Section 3.1] The formal definition says each rule must apply to 'strictly more than half' of cases; since case sets contain 5 to 10 cases, this is equivalent to a majority, but the wording could be simplified for readability.
[Dataset release] The paper does not state whether LRI-AUTO, LRI-GOLD, and the evaluation code will be publicly released; a data/code availability statement would improve reproducibility and is important for a benchmark paper.
[References] Several bibliographic entries are incomplete or inconsistent, such as 'Mistral AI. Ministral 8b instruct.' missing a year, and some entries for 'Deepmind' and 'Anthropic' lack full publication details.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The central claims rest on a domain-specific rule representation, a citation-based clustering assumption, hand-chosen support thresholds, and the reliability of model-generated training labels. No new physical or symbolic entities are postulated beyond the datasets themselves.

free parameters (4)

Majority support threshold tau = 50% (or > M/2, and >= 51% in prompts)
Defines which implicit rules are valid and is used in SILVER verification; chosen by hand, not derived from legal theory.
Case set size bounds = 5 to 10 cases
Filters and caps case set sizes to constrain input context and solution space; directly affects dataset composition.
LoRA rank and alpha = 8
Hyperparameters for fine-tuning small LLMs, chosen without reported tuning.
LoRA training epochs = 3
Fine-tuning schedule chosen by the authors; no epoch search is reported.

assumptions (4)

domain assumption Legal rules can be represented by the three-element structure (hypothetical condition, behavior pattern, legal consequence).
Adopted from Chinese jurisprudence (Wenxian et al., 2018); the entire task definition, annotation guidelines, and evaluation prompts depend on this structure.
domain assumption Cases citing identical statutory provisions form analogous precedent sets suitable for rule induction.
The clustering procedure (Section 4.1) assumes citation similarity implies enough commonality to induce shared rules.
ad hoc to paper Explicit rules must apply to 100% of cases and implicit rules to more than 50% of cases.
These thresholds are chosen in Section 3.1 and are not derived from legal theory; they shape both dataset filtering and SILVER verification.
domain assumption DeepSeek-R1's extractions are reliable enough to serve as training labels after filtering.
The LRI-AUTO training set is built entirely from DeepSeek-R1 outputs (Section 4.3) with only filter checks, so the quality of the training signal inherits that model's reliability.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Legal Rule Induction: Towards Generalizable Principle Discovery from Analogous Judicial Precedents." pith.science (2026). https://pith.science/paper/T4ZNXIFM

@misc{pith2026250514104,
  author       = {Pith},
  title        = {Pith review of: Legal Rule Induction: Towards Generalizable Principle Discovery from Analogous Judicial Precedents},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T4ZNXIFM}},
  note         = {Machine review of arXiv:2505.14104}
}

read the original abstract

Legal rules encompass not only codified statutes but also implicit adjudicatory principles derived from precedents that contain discretionary norms, social morality, and policy. While computational legal research has advanced in applying established rules to cases, inducing legal rules from judicial decisions remains understudied, constrained by limitations in model inference efficacy and symbolic reasoning capability. The advent of Large Language Models (LLMs) offers unprecedented opportunities for automating the extraction of such latent principles, yet progress is stymied by the absence of formal task definitions, benchmark datasets, and methodologies. To address this gap, we formalize Legal Rule Induction (LRI) as the task of deriving concise, generalizable doctrinal rules from sets of analogous precedents, distilling their shared preconditions, normative behaviors, and legal consequences. We introduce the first LRI benchmark, comprising 5,121 case sets (38,088 Chinese cases in total) for model tuning and 216 expert-annotated gold test sets. Experimental results reveal that: 1) State-of-the-art LLMs struggle with over-generalization and hallucination; 2) Training on our dataset markedly enhances LLMs capabilities in capturing nuanced rule patterns across similar cases.

Figures

Figures reproduced from arXiv: 2505.14104 by the authors.

**Figure 2.** The overview of the LRI-AUTO dataset curation pipeline (for civil and criminal cases) and main methods for rule induction, including LoRA, which utilizes LRI-AUTO for tuning and the LRI-GOLD dataset for testing. decompose the rule induction task into intermediate analytical stages (e.g., identifying common factual patterns, discerning judicial reasoning, and then formulating a rule). Long Chain-of-Thought Long-CoT r… view at source ↗

**Figure 3.** Distribution of rule set sizes across case num [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (12 more)

**Figure 4.** Figure 4: Scores (%) of different baselines. For the [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: Performance trends of Direct Induction of ten LLMs across varying case set sizes. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

**Figure 6.** Figure 6: Comparison of token usage (Input & Output) [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

**Figure 7.** Figure 7: Original case set size distribution before re-sampling. [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]

**Figure 8.** Figure 8: Case length distribution in LRI dataset. [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]

**Figure 9.** Figure 9: Rule number per case set distribution in LRI [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]

**Figure 10.** Figure 10: Performance trends of CoT of ten LLMs across varying case set sizes. [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]

**Figure 11.** Figure 11: Performance trends of Long-CoT of five LRMs across varying case set sizes. [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]

**Figure 12.** Figure 12: Performance trends of SILVER of five LLMs across varying case set sizes. [PITH_FULL_IMAGE:figures/full_fig_p016_12.png]

**Figure 13.** Figure 13: An example of a rule set generated by Llama-3.2-3B before and after training on the LRI-AUTO dataset [PITH_FULL_IMAGE:figures/full_fig_p026_13.png]

**Figure 14.** Figure 14: A criminal case from CJO after case processing. [PITH_FULL_IMAGE:figures/full_fig_p027_14.png]

**Figure 15.** Figure 15: A civil case from CJO after case processing. [PITH_FULL_IMAGE:figures/full_fig_p027_15.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 20 canonical work pages

[1]

may”, “is allowed to

Legal Rule Categories:Use only one of the follow- ing: (1) Criminal (2) Civil (3) Procedural (Litigation Procedure) 2.Legal Rule Structure:Each rule must include: (a) Hypothetical Condition– the context and sub- ject. (b)Behavior Pattern– classified as: • Permissive: “may”, “is allowed to”. • Obligatory: “must”, “shall”. • Prohibitive: “must not”, “is pro...

work page
[2]

may,” “is entitled to,

Behavioral Pattern:Specifies how people should act, including permissive, obligatory, and prohibitive patterns. -Permissive pattern:Uses expressions such as “may,” “is entitled to,” or “is allowed to.” -Obligatory pattern:Uses expressions such as “shall,” “must,” or “has the obligation to.” -Prohibitive pattern:Uses expressions such as “prohibited,” “shal...

work page
[3]

if the driver fails to exercise reasonable care, liability may be increased

Legal Consequence:Specifies the consequences of complying or not complying with the behavioral pattern. - Positive consequence: Legal effect resulting from compliance. - Negative consequence: Legal liability resulting from violation. Here are examples of the three behavioral patterns: 1.Permissive: Hypothetical Condition: A natural person wishes to engage...

work page
[6]

caused property loss to others

Output Format:Use JSON with keys: Explicit Rule, Implicit Rule. Table 4: Annotation guideline for legal rule induction that the number of rules per case set in LRI-GOLD is slightly lower than in LRI-AUTO. C Implementation Details C.1 Model Details In our experiments, we evaluate a total of 19 LLMs, categorized as follows: LLMsGPT-4o-mini(Hurst et al., 202...

work page 2024
[7]

Hypothetical Conditions:Conditions and circumstances under which the rule applies, including applicable subjects and their behaviors

work page
[10]

If [Hypothetical condition], then [behavioral pattern], then/otherwise [legal consequence]

Each rule must include all three components, forming a complete logical chain: “If [Hypothetical condition], then [behavioral pattern], then/otherwise [legal consequence].”

work page
[16]

Extracted Rules

Avoid duplication; merge similar rules. Table 10: Prompt of legal rule induction from a case set in the evaluation phase. The set of legal cases is as follows: {Legal Case Set} [Output Format] Please output in the following JSON format: { "Extracted Rules": [ { "Type": "Criminal", "Content": { "Hypothetical Condition": "A natural person with full criminal...

work page
[17]

If [Hypothetical condition], and [behavioral pattern], then/otherwise [legal consequence]

Each rule must include all three components, forming a complete logical chain: “If [Hypothetical condition], and [behavioral pattern], then/otherwise [legal consequence].”

work page

Show all 22 references

[18]

Do not use specific article numbers, content, or charges; summarize into a general rule applicable to the given case set
[19]

Must be inferred from commonalities among cases and should reflect discretionary standards in judicial practice
[20]

if the driver fails to exercise reasonable care, liability may be increased

The extracted rules must apply to≥51% of the cases. Example: Infer from all traffic accident cases in the set that “if the driver fails to exercise reasonable care, liability may be increased.”
[21]

Combining multiple relevant provisions to form a composite rule is allowed
[22]

Type annotation: Criminal / Civil / Procedural; do not use other types
[23]

Extracted Rules

Avoid duplication; merge similar rules. Table 12: Prompt of new rule induction. The set of legal cases is as follows: {Legal Case Set} The rules already extracted are as follows, please do not extract them again: {Already Extracted Rules} Please do not extract existing rules a...
[24]

•Behavioral Pattern:Whether the type (may do/should do/must not do) is accurately marked and described

Three-element check •Hypothetical Condition:Whether the preconditions for rule application are clearly defined. •Behavioral Pattern:Whether the type (may do/should do/must not do) is accurately marked and described. • Legal consequences:Whether it includes the positive and neg...
[25]

Prohibited content check • Whether there are prohibited references such as legal article numbers, names of crimes, etc
[26]

• Whether it is logically equivalent to any rule in the implicit rule set

Rule coverage check • Whether it is logically equivalent to any rule in the explicit rule set. • Whether it is logically equivalent to any rule in the implicit rule set
[27]

Correct". • In the coverage check,

Assessment conclusion • Rules that meet all the above requirements are"Correct". • In the coverage check,"logical equivalence"must be achieved to be considered"Correct". • If it does not meet the three-element check or contains prohibited content, it is"Incorrect". • If it doe...

1978
[2009]

InPro- ceedings of the 12th International Conference on Arti- ficial Intelligence and Law, ICAIL ’09, page 98–107, New York, NY , USA

Argumentation mining: the detection, clas- sification and structure of arguments in text. InPro- ceedings of the 12th International Conference on Arti- ficial Intelligence and Law, ICAIL ’09, page 98–107, New York, NY , USA. Association for Computing Machinery. Seth Polsley, P...
[2016]

InProceedings of COLING 2016, the 26th International Conference on Computational Linguistics: System Demonstra- tions, pages 258–262, Osaka, Japan

Casesummarizer: A system for automated summarization of legal texts. InProceedings of COLING 2016, the 26th International Conference on Computational Linguistics: System Demonstra- tions, pages 258–262, Osaka, Japan. The COLING 2016 Organizing Committee. Prakash Poudyal, Jarom...

2016 arXiv
[2020]

three-element theory

Legal knowledge extraction for knowledge graph based question-answering. InLegal knowledge and information systems, pages 143–153. IOS Press. Joshua B Tenenbaum, Charles Kemp, Thomas L Grif- fiths, and Noah D Goodman. 2011. How to grow a mind: Statistics, structure, and abstra...

2011 arXiv
[2024]

Brenden M Lake, Ruslan Salakhutdinov, and Joshua B Tenenbaum

Openai o1 system card.arXiv preprint arXiv:2412.16720. Brenden M Lake, Ruslan Salakhutdinov, and Joshua B Tenenbaum. 2015. Human-level concept learning through probabilistic program induction.Science, 350(6266):1332–1338. Grant Lamond. 2005. Do precedents create rules?Le- gal ...

2015 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

may”, “is allowed to

Legal Rule Categories:Use only one of the follow- ing: (1) Criminal (2) Civil (3) Procedural (Litigation Procedure) 2.Legal Rule Structure:Each rule must include: (a) Hypothetical Condition– the context and sub- ject. (b)Behavior Pattern– classified as: • Permissive: “may”, “is allowed to”. • Obligatory: “must”, “shall”. • Prohibitive: “must not”, “is pro...

work page

[2] [2]

may,” “is entitled to,

Behavioral Pattern:Specifies how people should act, including permissive, obligatory, and prohibitive patterns. -Permissive pattern:Uses expressions such as “may,” “is entitled to,” or “is allowed to.” -Obligatory pattern:Uses expressions such as “shall,” “must,” or “has the obligation to.” -Prohibitive pattern:Uses expressions such as “prohibited,” “shal...

work page

[3] [3]

if the driver fails to exercise reasonable care, liability may be increased

Legal Consequence:Specifies the consequences of complying or not complying with the behavioral pattern. - Positive consequence: Legal effect resulting from compliance. - Negative consequence: Legal liability resulting from violation. Here are examples of the three behavioral patterns: 1.Permissive: Hypothetical Condition: A natural person wishes to engage...

work page

[4] [6]

caused property loss to others

Output Format:Use JSON with keys: Explicit Rule, Implicit Rule. Table 4: Annotation guideline for legal rule induction that the number of rules per case set in LRI-GOLD is slightly lower than in LRI-AUTO. C Implementation Details C.1 Model Details In our experiments, we evaluate a total of 19 LLMs, categorized as follows: LLMsGPT-4o-mini(Hurst et al., 202...

work page 2024

[5] [7]

Hypothetical Conditions:Conditions and circumstances under which the rule applies, including applicable subjects and their behaviors

work page

[6] [10]

If [Hypothetical condition], then [behavioral pattern], then/otherwise [legal consequence]

Each rule must include all three components, forming a complete logical chain: “If [Hypothetical condition], then [behavioral pattern], then/otherwise [legal consequence].”

work page

[7] [16]

Extracted Rules

Avoid duplication; merge similar rules. Table 10: Prompt of legal rule induction from a case set in the evaluation phase. The set of legal cases is as follows: {Legal Case Set} [Output Format] Please output in the following JSON format: { "Extracted Rules": [ { "Type": "Criminal", "Content": { "Hypothetical Condition": "A natural person with full criminal...

work page

[8] [17]

If [Hypothetical condition], and [behavioral pattern], then/otherwise [legal consequence]

Each rule must include all three components, forming a complete logical chain: “If [Hypothetical condition], and [behavioral pattern], then/otherwise [legal consequence].”

work page