Research and Claim Checking
What is Evidence Verification Mode?
Evidence Verification Mode guides research tasks through testable claims, source minimums, and explicit unknowns.
Core Idea
In Short
Evidence Verification Mode guides research tasks through testable claims, source minimums, and explicit unknowns.
What belongs here?
- Turns factual and research tasks into testable claims.
- Requires stronger evidence for critical domains and marks unsupported claims as unknown.
- Keeps research verification separate from platform learning and UI automation.
When EVM applies
EVM applies to factual or research-based tasks: prices, deadlines, legal or medical claims, security questions, political or current information. UI automation such as clicking, typing, or navigating is not an EVM case.
How the agent answers
The agent returns results, sources, and unknowns. If a claim cannot be supported, it gives one useful next verification step instead of hiding a guess.
Memory connection
Verified claims can be recorded as learning candidates. Research preferences or reusable rules belong in Operator Notes.
Technical Data
| Element | Explanation |
|---|---|
structuredContent.evm |
Agent-readable block in task instructions with rules for source minimums, unknowns, and critical domains. It is not active in Learn Mode. |
criticalSourceMinimum |
Critical claims require more independent sources than simple non-critical facts. |
componentFilter="evm" |
Agents can inspect EVM-related learning statistics separately. This is an integrator surface, not a normal UI command. |
| MCP Variables | Explanation |
|---|---|
nova.get_instructions |
Read Learn Mode instructions. Supports the normal workflow in this area. |
nova.memory_add_candidate |
Record an explicit learning candidate. Supports the normal workflow in this area. |
nova.memory_stats |
Review learning-memory statistics. Reads current state and makes visible what is happening. |
nova.operator_notes_store |
Store an operator note. Starts, changes, or controls a bounded step. |
nova.operator_notes_query |
Search operator notes. Starts, changes, or controls a bounded step. |
Related Areas
- Operator Notes
- Operator Notes können Recherchepräferenzen für spätere EVM-Aufgaben halten.
- LCJ
- LCJ kann verifizierte oder unbekannte Claims als Kandidaten halten.
- Learning
- EVM ist angrenzend zu Lernen, aber keine eigene Lernsäule.