authority-aware-structural-re-ranking

name

authority-aware structural re-ranking

description

Use this skill when you need to find the 'most important' or 'landmark' documents in a collection where popularity and rank matter as much as keywords. Trigger it for requests like 'find the landmark rulings', 'give me the most cited research papers', 'prioritize high-court decisions over local ones', or 'rank the results by how much other documents point to them'. This skill is essential for navigating legal, academic, or professional corpora where information is organized in a hierarchy or linked through a citation network.

Skill: authority-aware structural re-ranking

1. Capability Definition & Real Case

• Professional Definition: The ability to prioritize retrieved documents by integrating networked authority metrics (such as PageRank centrality) and institutional hierarchies (such as court tiers or organizational ranks) with traditional semantic similarity. This capability ensures the Agent identifies doctrinally authoritative sources rather than just lexically similar ones, mitigating 'naive retrieval' errors in specialized knowledge domains.
• Dimension Hierarchy: Specialized Knowledge Navigation->Corpus-Specific Retrieval & Reasoning->authority-aware structural re-ranking

Real Case

[Case 1]

• Initial Environment: A RAG environment containing a citation-weighted knowledge graph of U.S. copyright law precedents, including metadata for court hierarchies (Supreme Court, Appellate, District) and inter-case citations.
• Real Question: Analyze the 'amount and substantiality' factor for a parody case where a song heavily samples a 1960s pop hit.
• Real Trajectory: 1. Search for 'parody sampling' in the vector database. 2. Identify two similar cases: one recent obscure District Court ruling and one older highly-cited Supreme Court case. 3. Observe the PageRank score: The Supreme Court case (Campbell v. Acuff-Rose) has a disproportionality high ranking compared to the local ruling. 4. Re-rank the results to prioritize the landmark case. 5. Extract the 'Factor 3' analysis from the Supreme Court text to ground the answer.
• Real Answer: According to the Supreme Court's ruling in Campbell v. Acuff-Rose Music, Inc., even if a large portion is used, it may be fair if the 'heart' is necessary to conjure up the original for parody purposes.
• Why this demonstrates the capability: The agent avoids 'naive retrieval' by prioritizing the doctrinally authoritative source. Instead of following the most 'recent' or 'lexically close' local ruling, it uses the citation network (PageRank) and hierarchy (Supreme Court) to identify the landmark precedent that actually governs the doctrine.

---

[Case 2]

• Initial Environment: A technical repository of global engineering standards (ISO, IEEE) where 'Base Standards' are linked to 'Regional Amendments' and 'Draft Proposals' via a hierarchy of authority.
• Real Question: What is the mandatory safety clearance for high-voltage server racks in industrial zones?
• Real Trajectory: 1. Retrieve 10 standard documents mentioning 'server rack clearance'. 2. Detect a conflict between a 2024 'Draft Proposal' and the 2021 'Base Standard'. 3. Analyze the institutional hierarchy: The Base Standard is the authoritative parent, while the Draft is a lower-rank amendment. 4. Weight the retrieval result based on the 'Standard Status' metadata. 5. Provide the grounded answer based on the high-rank mandatory code.
• Real Answer: The mandatory clearance is 3 feet as specified in the authoritative IEEE Base Standard, despite lower-rank draft proposals suggesting a reduction.
• Why this demonstrates the capability: This case tests the ability to resolve informational conflicts by using structural rank. The agent recognizes that a lower-authority 'Draft' cannot override the institutional 'Base', maintaining factual reliability through hierarchical grounding.

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:

1. Phase 1: Environment Exploration Read the exploration guidelines to discover raw knowledge seeds: references/EXPLORATION.md

2. Phase 2: Trajectory Selection Once Phase 1 is complete, read the selection criteria to evaluate the trajectory: references/SELECTION.md

3. Phase 3: Data Synthesis Once a trajectory passes Phase 2, read the synthesis instructions to generate the final data: references/SYNTHESIS.md