# tinker_cookbook.exceptions.EvalError

## *class* [**tinker_cookbook.exceptions.EvalError**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/exceptions.py#L221)(*[TinkerCookbookError](https://tinker-docs.thinkingmachines.ai/cookbook/api-reference/exceptions/tinkercookbookerror/index.md), RuntimeError*)

An error during benchmark evaluation.

Base class for eval-related errors. Inherits from `RuntimeError` because eval errors are operational failures (network, timeout, sandbox) rather than configuration issues.

Eval errors are **non-fatal by default**: the benchmark runner catches them per-example, records them as scored failures (reward=0), and continues with remaining examples. The error details are preserved in `StoredTrajectory.error` for post-hoc analysis.

## Referenced by

- [tinker_cookbook.exceptions.BenchmarkNotFoundError](https://tinker-docs.thinkingmachines.ai/cookbook/api-reference/exceptions/benchmarknotfounderror/index.md)
- [tinker_cookbook.exceptions.EvalGradingError](https://tinker-docs.thinkingmachines.ai/cookbook/api-reference/exceptions/evalgradingerror/index.md)
- [tinker_cookbook.exceptions.EvalTimeoutError](https://tinker-docs.thinkingmachines.ai/cookbook/api-reference/exceptions/evaltimeouterror/index.md)