RECORDED RESULTS
One study. Two useful views.
- Evaluation scenarios
- 500
- Selected training examples
- 447
One model and a fixed recovery protocol. These results describe this study, not general agent safety.
Explore the outcome breakdownGOLD TRACE DATAWORKS
Gold Trace Dataworks builds AI agents, evaluates how they use tools, and turns recorded runs into datasets with traceable sources and labels.
Custom workflows, tool connections, and MCP integrations.
Defined scenarios, versioned criteria, and recorded outcomes.
Structured examples with source history and documented selection.
SELECTED WORK / TOOL FAILURE RECOVERY
In a 500-scenario evaluation, 447 scenarios met the recovery criteria. The other 53 exhausted their retry budget.
We kept both outcomes. The complete evaluation and the selected training examples can be traced back to the same recorded runs.
RECORDED RESULTS
One model and a fixed recovery protocol. These results describe this study, not general agent safety.
Explore the outcome breakdownBring us an agent to build, a workflow to test, or data that needs a clearer evaluation standard.