| name | reasoning-intensive analogical retrieval |
| description | Use this skill when the user wants to test the agent's ability to find relevant documents that do not share many words with the question but are linked by deeper logic. Trigger this for requests like 'find a case that matches this story', 'give me law that applies to this situation even if the terms are different', 'search based on the legal issue rather than keywords', or 'test if the model can spot the underlying rule in a complex scenario'. It is specifically designed for high-reasoning legal or technical retrieval where lexical overlap is extremely low. |
Skill: reasoning-intensive analogical retrieval
1. Capability Definition & Real Case
- Professional Definition: The ability to retrieve relevant evidence from a bounded corpus when the query consists of a complex, specific factual narrative (e.g., a hypothetical scenario) and the target document consists of abstract principles or higher-order rules (e.g., judicial opinions or statutes). This capability focuses on bridging the 'lexical gap'—where query and document share low keyword similarity—by performing latent issue spotting and analogical reasoning during the retrieval process.
- Dimension Hierarchy: Specialized Knowledge Navigation->Corpus-Specific Retrieval & Reasoning->reasoning-intensive analogical retrieval
Real Case
[Case 1]
- Initial Environment: A large retrieval corpus containing approximately 900,000 document segments, including United States federal caselaw paragraphs and legal encyclopedia entries.
- Real Question: A teacher at a private school receiving 20% of its budget from state grants and tax exemptions is fired for a lecture criticizing school uniforms. If the teacher sues in federal court alleging a violation of constitutional rights, what is the likely result?
- Real Trajectory: 1. Identify the 'State Action' issue: The school is private, but receives state aid. 2. Formulate an expansion query focused on whether private school personnel matters constitute state action. 3. Retrieve documents regarding the 14th Amendment and the distinction between 'exclusively public functions' and 'private conduct'. 4. Match the specific case evidence (20% funding, teacher firing) to the precedent regarding state involvement in personnel matters.
- Real Answer: The claim fails because assistance and involvement by the state did not result in the private school's action being conduct by the state.
- Why this demonstrates the capability: The question describes a specific story about uniforms and grants, while the correct document addresses the abstract 'State Action Doctrine'. There is almost no lexical overlap between 'uniform protest' and '14th Amendment personnel matters', forcing the agent to reason about the legal category rather than simply match terms.
[Case 2]
- Initial Environment: A bounded database of housing statutes and eviction laws across multiple jurisdictions (~2 million passages).
- Real Question: Does the law in Alabama specify rebuttals or specific defenses available to tenants subject to eviction proceedings?
- Real Trajectory: 1. Target the jurisdiction (Alabama) and the domain (housing/eviction). 2. Expand search keywords from 'rebuttals' to include 'counterclaims', 'defenses', and 'retaliatory action'. 3. Locate statutory sections such as 35-9A-407 that discuss tenant remedies. 4. Verify that the retrieved statutes explicitly mention a 'defense' in retaliatory actions.
- Real Answer: Yes
- Why this demonstrates the capability: The question uses the word 'rebuttal', a common layman or broad legal term, but the statutory documents use highly specific terms like 'remedies', 'Section 35-9A-407', and 'counterclaim'. The agent must bridge the semantic gap between a general inquiry and precise statutory language.
Pipeline Execution Instructions
To synthesize data for this capability, you must strictly follow a 3-phase pipeline. Do not hallucinate steps. Read the corresponding reference file for each phase sequentially:
-
Phase 1: Environment Exploration
Read the exploration guidelines to discover raw knowledge seeds:
references/EXPLORATION.md
-
Phase 2: Trajectory Selection
Once Phase 1 is complete, read the selection criteria to evaluate the trajectory:
references/SELECTION.md
-
Phase 3: Data Synthesis
Once a trajectory passes Phase 2, read the synthesis instructions to generate the final data:
references/SYNTHESIS.md
