case-officer

Case Officer Agent

Investigation synthesis agent that reads sources, generates hypotheses, evaluates evidence, and produces actionable intelligence reports with next steps.

What It Does

The Case Officer is the second phase of Archive Research. It answers: "What can be concluded, and what cannot—based on evidence and constraints?"

flowchart LR
    A[Quartermaster Intel] --> B[Read Sources]
    B --> C[Generate Hypotheses]
    C --> D[Evaluate Evidence]
    D --> E[Expand if Needed]
    E --> F[Synthesize Report]
    F --> G[Next Steps]

Key capabilities:

• Source Reading with multi-reader cascade (Perplexity, Jina, AgentQL, HTTP)
• Context Budget Management - Tracks token usage across document reads
• Hypothesis Generation using DSPy with evidence tracking
• Evidence Evaluation - SUPPORTS, REFUTES, or NEUTRAL for each hypothesis
• Autonomous Expansion - Searches for more evidence when needed
• Model Escalation - Progressively uses more powerful models on retry
• PDF Intelligence with Aryn context-aware extraction

Use When

• Quartermaster has mapped archive sources
• You need to synthesize findings into a coherent investigation
• You want structured hypotheses with evidence evaluation
• You need actionable next steps with access instructions

Prerequisites

• Quartermaster results in tree context (runs automatically)
• Document reader configured (at least one):
  • PERPLEXITY_API_KEY (recommended - reads any size)
  • JINA_API_KEY (fast, good for articles)
  • AGENTQL_API_KEY (JS-heavy sites)
• Optional: ARYN_API_KEY for intelligent PDF preview

Architecture

Core Components

backend/elysia/tools/archives/
├── case_officer_tool.py       # Main agent logic
├── dspy_programs.py           # DSPy signatures and programs
└── types.py                   # Hypothesis, CaseOfficerResult

backend/elysia/api/services/
├── document_reader_service.py # Multi-reader cascade
└── sofia_service.py           # Search for expansion