Monolithic Dynamic Pipeline Internal Flow

flowchart TB
    subgraph Entry[Entry Point]
        INPUT[runMonolithicDynamic]
        BUDGET[Initialize Budget maxIterations 4 maxSearches 6]
        TIMEOUT[Start timeout 2.5 minutes]
    end

    subgraph Analysis[Single LLM Analysis]
        PROMPT[Build dynamic analysis prompt]
        LLM[LLM Call DynamicAnalysisSchema]
        SEARCH[searchWebWithProvider]
        FETCH[fetchSourceContent]
    end

    subgraph Safety[Safety Contract Enforcement]
        CITATIONS[Extract citations array]
        RAW[Store rawJson]
        VALIDATE[Validate minimum fields]
    end

    subgraph Transform[Output Transform]
        DYNAMIC[Build DynamicAnalysisResult]
        REPORT[Generate experimental report]
    end

    subgraph Output[Output]
        RESULT[resultJson dynamic structure]
        MARKDOWN[reportMarkdown experimental label]
    end

    INPUT --> BUDGET
    BUDGET --> TIMEOUT
    TIMEOUT --> PROMPT

    PROMPT --> LLM
    LLM --> SEARCH
    SEARCH --> FETCH
    FETCH --> LLM

    LLM --> CITATIONS
    CITATIONS --> RAW
    RAW --> VALIDATE

    VALIDATE --> DYNAMIC
    DYNAMIC --> REPORT
    REPORT --> RESULT
    REPORT --> MARKDOWN

Key Characteristics

Budget Configuration

Minimum Safety Contract

Required fields (always present):

Optional fields (LLM decides):

Output Schema

erDiagram
    DYNAMIC_RESULT ||--o{ CITATION : has
    DYNAMIC_RESULT ||--o| VERDICT : may_have
    DYNAMIC_RESULT ||--o{ FINDING : may_have

    DYNAMIC_RESULT {
        array citations required
        any rawJson required
        string summary
        string methodology
        array limitations
    }

    CITATION {
        string url required
        string title
        string excerpt required
        string accessedAt
    }

    VERDICT {
        string label
        number score 0_to_100
        number confidence 0_to_100
        string reasoning
    }

    FINDING {
        string point
        string support strong_moderate_weak_none
        array sources
        string notes
    }

Differences from Other Pipelines

Use Cases

When to use Dynamic:

• Exploratory analysis of novel claim types
• Research where canonical structure is limiting
• Comparing LLM reasoning approaches
• Debugging LLM behavior

When NOT to use:

• Production fact-checking
• Results that need UI rendering
• Comparing results across pipelines