dingxingdi Skill: cross-part comparative reasoning
1. Capability Definition & Real Case
- Professional Definition: The ability to synthesize, compare, and contrast information distributed across disparate parts of a long context or multiple documents. Unlike simple integration, this capability requires the agent to identify discrete comparable 'anchors' (such as dates, entities, or theories), retrieve the specific attributes associated with each anchor, and generate a response that articulates the similarities, differences, or changes between them.
- Dimension Hierarchy: Multi-Source Evidence Composition->Cross-Document Synthesis->cross-part comparative reasoning
Real Case
[Case 1]
- Initial Environment: A RAG environment containing a financial statement for CISO Global, Inc., including balance sheets from December 31, 2023, and March 31, 2024.
- Real Question: How did the total current assets and total current liabilities change for CISO Global, Inc. from December 31, 2023, to March 31, 2024?
- Real Trajectory: 1. Search for 'total current assets' and 'total current liabilities' specifically for the date December 31, 2023. 2. Search for the same metrics specifically for March 31, 2024. 3. Extract the assets ($10,957,814 vs $9,276,063) and liabilities ($26,071,102 vs $32,604,126) from the two separate report sections. 4. Synthesize the comparison by identifying the direction of change for both metrics.
- Real Answer: CISO Global’s total current assets decreased from $10,957,814 on December 31, 2023, to $9,276,063 on March 31, 2024, while total current liabilities increased from $26,071,102 to $32,604,126 in the same period.
- Why this demonstrates the capability: The question requires information from two distinct chronological snapshots within a financial statement. The agent cannot answer by finding one single sentence; it must locate two separate data points (assets and liabilities) for two distinct timeframes and articulate the delta between them. This demonstrates the capability to handle cross-part comparative reasoning over quantitative data.
[Case 2]
- Initial Environment: A RAG environment containing multiple academic papers related to Information Retrieval and document ranking.
- Real Question: How do the approaches for incorporating context into ranking functions differ between the method proposed in paper 0 and in paper 1?
- Real Trajectory: 1. Identify 'paper 0' and 'paper 1' in the corpus. 2. Retrieve the 'Methods' or 'Proposed Approach' section for paper 0 to find current context mechanisms (delta features). 3. Retrieve the corresponding section for paper 1 (self-attention mechanism). 4. Contrast the two mechanisms by highlighting the specific technical difference (neighbors vs self-attention).
- Real Answer: Paper 0 incorporates context using delta features comparing neighboring items, while paper 1 uses a self-attention mechanism to account for interactions between items during both training and inference.
- Why this demonstrates the capability: This case requires cross-document comparative reasoning across technical literature. The agent must precisely identify the specific methodology from two different 'poles' (the two papers) and formulate a contrasting statement. Success depends on isolating the unique technical contribution of each document relative to the same topic (contextual ranking).
[Case 3]
- Initial Environment: A RAG environment containing the full text of a classical novel featuring multiple characters and a narrator.
- Real Question: How does the narrator’s attitude towards Turkey and Nippers compare to his attitude towards Bartleby?
- Real Answer: The narrator is willing to overlook Turkey’s and Nippers’ flaws due to their usefulness, just as he tolerates Bartleby’s peculiar behavior because of his steady work and presence.
- Why this demonstrates the capability: This task involves subjective, qualitative comparison across a narrative context. The agent must scan multiple character descriptions and internal monologues spread across the book to extract the narrator's specific sentiments toward different entities. The capability is demonstrated by finding the common theme (tolerance/utility) that links the disparate character evaluations.
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