{
  "contract_id": "f8-c04-feature-contract-v1",
  "owner": "equipo-datos-ia",
  "purpose": "Construir features tabulares y representaciones vectoriales para soporte académico sin usar columnas de identidad o target como entrada del modelo.",
  "entity_key": "case_id",
  "event_time": "created_at",
  "split_file": "data/split_assignments.csv",
  "target": "label",
  "allowed_input_columns": [
    "created_at",
    "product",
    "channel",
    "text"
  ],
  "forbidden_input_columns": [
    "case_id",
    "student_id",
    "source_id",
    "label"
  ],
  "fit_scope": "train",
  "transform_scope": [
    "train",
    "validation",
    "test"
  ],
  "categorical_features": [
    "product",
    "channel"
  ],
  "numeric_features": [
    "days_since_min_train_date",
    "text_token_count"
  ],
  "text_feature": {
    "column": "text",
    "method": "train_tfidf",
    "max_terms": 40,
    "min_token_length": 3
  },
  "dense_embedding": {
    "method": "deterministic_hash_projection",
    "dimensions": 64,
    "normalization": "l2",
    "purpose": "Baseline local para probar contrato, dimensiones y similitud antes de sustituirlo por un encoder neural."
  },
  "similarity": {
    "metric": "cosine",
    "top_k": 3,
    "index_splits": [
      "train"
    ]
  },
  "quality_gates": {
    "max_unknown_categories": 0,
    "max_forbidden_features_used": 0,
    "require_feature_manifest": true,
    "require_l2_normalized_dense_vectors": true
  }
}