dingxingdi Skill: scientific threshold and context inference
1. Capability Definition & Real Case
- Professional Definition: The capability to operationalize expert-level domain expressions, persona-centric business logic, and analytical workflows by mapping them to specific relational implementations. This involves interpreting unstated business constraints, applying hierarchical workflow logic (persona → scenario → workflow), and retrieving implicit business rules or ontological constants from reference tables to bridge the gap between casual professional intent and database-level calculations.
- Dimension Hierarchy: Query Reasoning->Domain-Constrained Semantics->scientific threshold and context inference
Real Case
[Case 1]
- Initial Environment: A production-scale Salesforce-like CRM database containing 'Leads', 'Opportunities', and 'User' tables. The schema follows a standard B2B sales data model but does not explicitly define 'Conversion Rate' as a column.
- Real Question: For this month’s pipeline health review, what is the MQL-to-SQL conversion rate for the Sales Development team?
- Real Trajectory: 1. Identify the persona as a Sales Manager and the scenario as a pipeline review. 2. Map the domain term 'MQL' (Marketing Qualified Lead) to leads with status 'Qualified'. 3. Map 'SQL' (Sales Qualified Lead) to the corresponding stage in the Opportunity table. 4. Perform a join on the lead-opportunity conversion link. 5. Generate a ratio calculation (SUM(SQLs)/SUM(MQLs)) while applying a temporal filter for the current month.
- Real Answer: SELECT SUM(CASE WHEN T2.StageName = 'Sales Qualified' THEN 1 ELSE 0 END) * 1.0 / COUNT(T1.Id) as conversion_rate FROM Leads AS T1 LEFT JOIN Opportunities AS T2 ON T1.ConvertedOpportunityId = T2.Id WHERE T1.Status = 'Qualified' AND T1.CreatedDate = THIS_MONTH
- Why this demonstrates the capability: This case demonstrates business logic operationalization. The agent must infer the multi-step 'funnel' logic and unstated status definitions (MQL vs SQL) required for a standard sales workflow, effectively translating a high-level manager's goal into a precise mathematical ratio across joined tables.
[Case 2]
- Initial Environment: A financial database containing an 'Income' table and a 'User Profile' metadata store defining organizational report standards.
- Real Question: What is the revenue last month?
- Real Trajectory: 1. Consult the knowledge base for 'Revenue'; discover a policy stating 'Revenue queries for management must include YoY as the default metric'. 2. Retrieve the user profile 'Regional Manager' and append their specific implicit department filter, automatically calculating current vs prior year sums.
- Real Answer: SELECT SUM(ptd_amt) AS current, SUM(py_ptd_amt) AS prior, (SUM(ptd_amt) - SUM(py_ptd_amt))/SUM(py_ptd_amt) AS yoy FROM income_table WHERE period = '202504' AND dept = 'Sales';
- Why this demonstrates the capability: The agent infers an implicit unstated reporting policy. It operationalizes business defaults (YoY metric) and profile filtration policies not explicitly requested in the casual natural language text.
[Case 3]
- Initial Environment: A clinical database (MIMIC-III) where 'chartevents' stores measurements by 'itemid' without descriptive names. A reference table 'd_items' relates 'itemid' to the strings.
- Real Question: Find the average heart rate for patients in the last 24 hours.
- Real Trajectory: 1. Identify 'heart rate' as an unmapped entity. Search 'd_items' for the exact 'itemid' for Heart Rate. 2. Extract the code (e.g., 211) and inject it into the relational WHERE parameter targeting the 'chartevents' measurement table.
- Real Answer: SELECT AVG(valuenum) FROM chartevents WHERE itemid = 211 AND charttime > NOW() - INTERVAL '24 hours';
- Why this demonstrates the capability: This illustrates ontological value mapping as a domain inference step. The agent utilizes a structural reference table to map the high-level domain term into its required numerical ID before filtering.
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