| # Evaluation protocol | |
| Status: pending. This resource repository does not yet publish a finalized cross-model evaluation protocol, benchmark split or new benchmark scores. Per-model smoke runs and file-format checks are documented in model directories, but they are not a common quality comparison. | |
| The maintainer should specify task definitions, sample selection, input/output pairing, seeds, truncation, postprocessing, coordinate and robot conventions, resampling, metrics and aggregation. Record failures and skipped samples separately. Distinguish planned task coverage from completed experiments. | |
| See the [model information template](../templates/MODEL_INFORMATION_zh.md) and [validation template](../templates/VALIDATION_REPORT_zh.md). | |