Evidence · DiggingBeagle record
TraceTrial-Bench: executable correctness criteria for AI-agent harnesses
TraceTrial-Bench: executable correctness criteria for AI-agent harnesses
- Source role
- primary disclosure
Evidence record
TraceTrial-Bench: executable correctness criteria for AI-agent harnesses
Claim-level citations (2)
- supportsTraceTrial separates task completion from agent harness correctness: TraceTrial-Bench defines seven separate harness dimensions, including context, governance, lifecycle and verification, and uses execution traces to attribute failure rather than equating a completed task with an authorized workflow.
Abstract, seven components and trajectory tracing evaluation
- supportsTraceTrial separates task completion from agent harness correctness: TraceTrial-Bench evaluated 3,420 runs across 95 executable tasks, four harnesses and nine models. The authors report 31.1 percent task success combined with failure of at least one harness criterion.
Abstract; harness criteria and 3,420-trajectory evaluation