AgentTriage evidence-grounded tool-effect tracing across five runtimes

Evidence record

Read the original source ↗

Claim-level citations (2)

• supportsAgentTriage required runtime evidence before calling an agent action completed: The AgentTriage study describes a replayable diagnosis contract across five agent runtimes with groundable claims and mandatory abstention when evidence is incomplete.

  Abstract; five-runtime replay; diagnosis evidence constraints

• supportsAgentTriage required runtime evidence before calling an agent action completed: AgentTriage reports 159 applicable executions across five runtimes; on 120 adjudicated cases it measured grounded joint diagnosis of 0.858, required-evidence coverage of 0.966 and unsupported-claim rate of 0.036.

  Abstract; five-runtime normalized executions and 120 adjudicated cases; accuracy and coverage metrics