robust-multi-hop-path-recovery

# Skill: robust-multi-hop-path-recovery ## 1. Capability Definition & Real Case * **Professional Definition**: The ability to withstand knowledge incompleteness in relational or graph-structured corpora by identifying and traversing alternative multi-hop reasoning paths. This capability involves pivoting through intermediate entities or bridge nodes to reconstruct a fact when the primary direct evidence (single-hop relationship) is absent, ensuring the RAG system remains functional even under partial data loss. * **Dimension Hierarchy**: Specialized Knowledge Navigation->Corpus-Specific Retrieval & Reasoning->robust-multi-hop-path-recovery ### Real Case **[Case 1]** * **Initial Environment**: A relational knowledge base containing biographical data for celebrities. The direct link 'X is the brother of Y' has been removed from the database records for specific individuals. * **Real Question**: Who is the brother of Justin Bieber? * **Real Trajectory**: 1. Search for a direct 'brother' relationship for the entity Justin Bieber. 2. Observe that the 'has_brother' record is missing. 3. Pivot to a broader search of family relations. 4. Identify 'Jeremy Bieber' as the parent of Justin Bieber. 5. Search for other children of Jeremy Bieber in the corpus. 6. Identify 'Jaxon Bieber' as a child of Jeremy. 7. Synthesize the inference that Jaxon is the brother based on the shared parentage path. * **Real Answer**: Jaxon Bieber * **Why this demonstrates the capability**: This case demonstrates robustness to knowledge incompleteness. The agent cannot find the direct link but successfully compensates by executing a multi-hop inference (Person -> Parent -> Child) to recover the missing fact. --- **[Case 2]** * **Initial Environment**: A geographical and administrative corpus where country-to-timezone links are partially missing, but city-level administrative divisions and their respective timezones are preserved. * **Real Question**: What is the Nigeria time? * **Real Trajectory**: 1. Search for direct timezone attributes for 'Nigeria'. 2. Observation: The direct link is absent in the current knowledge fragment. 3. Redirect search to look for administrative divisions or major cities within Nigeria. 4. Retrieve 'Bauchi' as an administrative division of Nigeria. 5. Retrieve the timezone for 'Bauchi', which is 'West Africa Time Zone'. 6. Conclude that the country shares this timezone. * **Real Answer**: West Africa Time Zone * **Why this demonstrates the capability**: The agent overcomes 'Reasoning Path Disruption' by traversing an indirect route (Country -> Administrative Division -> Time Zone) when the direct country-level attribute is unavailable. --- **[Case 3]** * **Initial Environment**: A university directory where the direct mapping of a university to its home city is missing, but mappings between the university, its individual campuses, and those campuses' locations are available. * **Real Question**: What city is the University of Oregon in? * **Real Trajectory**: 1. Attempt to retrieve the city location for 'University of Oregon'. 2. Observation: Search returns no direct city field. 3. Search for related entities or subunits of the university. 4. Identify the 'Eugene Campus' as a primary unit. 5. Retrieve the location for the 'Eugene Campus', which is listed as 'Eugene'. 6. Logically link the campus location to the university's main city. * **Real Answer**: Eugene * **Why this demonstrates the capability**: This demonstrates path recovery via sub-entity attributes. The agent uses the campus as a semantic bridge to resolve a question that would otherwise fail due to an incomplete primary record. ## 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`