dingxingdi Skill: multi-conditional-entity-localization
1. Capability Definition & Real Case
- Professional Definition: The ability to pinpoint a unique entity or subject by logically intersecting multiple semantic constraints (triplets) distributed throughout a retrieved document pool. Unlike simple attribute integration—which aggregates data for a known entity—this capability requires the agent to perform 'subject identification' by uncovering an entity that simultaneously satisfies multiple independent predicates (e.g., Entity X has Relation 1 with A, and Entity X has Relation 2 with B) found across disjointed documents.
- Dimension Hierarchy: Multi-Source Evidence Composition->Cross-Document Synthesis->multi-conditional-entity-localization
Real Case
[Case 1]
- Initial Environment: A bounded news corpus containing factual snippets about Russian public figures, performances, and social events. The documents are split into small, non-overlapping chunks.
- Real Question: Who performed at M-bar and met with Dmitry Dibrov?
- Real Trajectory: 1. Analyze the two distinct constraints: 'performed at M-bar' and 'met with Dmitry Dibrov'. 2. Execute a search for 'Dmitry Dibrov' to find meeting records. 3. Retrieve a snippet identifying a meeting with 'Roman Miroshnichenko'. 4. Execute a secondary search for 'M-bar' performances. 5. Retrieve an independent snippet stating that 'Roman Miroshnichenko' performed at M-bar. 6. Intersect the two candidate sets: identify that 'Roman Miroshnichenko' is the only entity satisfying both the location constraint (M-bar) and the social constraint (Dmitry Dibrov).
- Real Answer: Roman Miroshnichenko
- Why this demonstrates the capability: This case demonstrates the 'Conditional' reasoning pattern where the answer is not a composite list of facts, but the identification of a specific entity that acts as the intersection of two distinct factual relations. The agent must verify the shared identity across disjointed contexts to localize the target.
[Case 2]
- Initial Environment: A RAG environment containing a series of daily automotive news updates and historical company country-of-origin records.
- Real Question: Which company, headquartered in China, sold exactly 2139 cars in the first quarter of 2023?
- Real Trajectory: 1. Decompose the query into two predicates: 'headquartered in China' and 'sold 2139 cars in 2023'. 2. Search for the specific sales figure '2139' within the database. 3. Retrieve a news update identifying 'FAW' as the company with that specific sales volume. 4. Pivot to verify the headquarters constraint for 'FAW'. 5. Retrieve a company profile document confirming FAW's country of origin is China. 6. Finalize the localize entity as FAW after meeting both conditions.
- Real Answer: FAW
- Why this demonstrates the capability: The question forces an intersection of a quantitative attribute (sales volume) and a qualitative metadata attribute (headquarters), where the entity itself (FAW) must be discovered rather than being provided as a keyword in the prompt. This requires filtering a candidate set through multiple logical gates.
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.mdPhase 2: Trajectory Selection Once Phase 1 is complete, read the selection criteria to evaluate the trajectory:
references/SELECTION.mdPhase 3: Data Synthesis Once a trajectory passes Phase 2, read the synthesis instructions to generate the final data:
references/SYNTHESIS.md