Evidence-Bounded Verification

Understanding AXIMUR's verification model and its boundaries.

What is Evidence-Bounded Verification?

Evidence-bounded verification means AXIMUR evaluates claims only against the specific evidence you provide. It does not:

• Search the web for additional information
• Query external knowledge bases or databases
• Access the LLM's training data
• Make open-world truth claims

Why This Matters

By being evidence-bounded, AXIMUR gives you:

• Predictability: You control the evidence source
• Transparency: You know exactly what was evaluated
• Auditability: Full audit trail links claim spans to source spans
• Safety: No hidden data sources or external influences

Verification Boundaries

Within Scope

• Evaluating claims against provided text sources
• Detecting contradictions between claim and source
• Flagging insufficient or ambiguous evidence
• Providing reasoning for the verdict

Outside Scope

• Web searching or document retrieval
• Determining absolute truth about the world
• Fact-checking without a provided source
• Learning from or adapting to your data

Practical Implications

Because AXIMUR is evidence-bounded:

• You are responsible for evidence quality — The verdict reflects your source, not ground truth
• Source selection matters — Choose relevant, accurate, up-to-date grounding sources
• Multiple sources help — Combine verification across multiple evidence sources for robustness
• UNCERTAIN is useful — It tells you the evidence is insufficient, prompting better retrieval

Limitations

See Limitations for complete details on constraints, rate limits, and operational boundaries.

Learn More

• Quickstart — Get started with the API
• RAG Claim Verification — Use in retrieval workflows
• Pre-Action Verification — Verify before agent actions
• FAQ — Common questions