name: run_evaluation
description: "Run a model evaluation on a dataset split and return performance metrics"
when_to_use: "When a model needs to be assessed on a held-out test set, comparing performance against baselines or prior checkpoints"
parameters:
  model_id:
    type: string
    description: "Identifier of the model to evaluate"
    required: true
  dataset_id:
    type: string
    description: "Identifier of the dataset to evaluate against"
    required: true
  split:
    type: string
    description: "Dataset split to use for evaluation"
    default: "test"
  metrics:
    type: array
    description: "List of metrics to compute, e.g. [\"accuracy\", \"f1\", \"auc\", \"rmse\"]"
    required: false
  batch_size:
    type: integer
    description: "Number of samples per evaluation batch"
    required: false
returns:
  type: object
  description: "Evaluation results including per-metric scores, timing, and sample count"
  properties:
    model_id:
      type: string
    dataset_id:
      type: string
    split:
      type: string
    metrics:
      type: object
      description: "Map of metric name to computed value, e.g. {accuracy: 0.91, f1: 0.88}"
    evaluation_time_s:
      type: number
    samples_evaluated:
      type: integer