Clinical Nutrition Instruction Data

Summary and intended purpose

Synthetic instruction-response records covering nutritional assessment, requirement estimation, and structured plan generation under stated clinical constraints. Records are designed to teach consistent structure and explicit constraint handling rather than to assert medical correctness.

Target training or evaluation tasks

• Nutritional requirement estimation from a stated scenario
• Structured plan generation with explicit constraint lists
• Identification of missing information before producing a recommendation
• Consistent JSON or schema-bound outputs

Clinical scope

• Adult clinical nutrition scenarios
• Guideline-mapped focus areas explored in internal R&D
• Common comorbidity combinations where constraints interact
• Out of scope: paediatric, acute inpatient, and non-nutrition specialties

Example record schema

{
  "example_id": "cni-0001-synthetic",
  "specialty": "clinical_nutrition",
  "task_type": "structured_plan_generation",
  "language": "en",
  "synthetic_patient": {
    "age_group": "adult_31_50y",
    "region": "US",
    "goals": [
      "weight_management"
    ],
    "dietary_pattern": "vegetarian",
    "measurements": {
      "bmi": 27.4,
      "activity_level": "sedentary"
    },
    "conditions": [
      "hypertension_stage_1"
    ]
  },
  "scenario": {
    "setting": "outpatient_followup",
    "summary": "Synthetic adult with stage 1 hypertension and a vegetarian diet, asking for a weight-management plan. Medication list and recent labs are not recorded.",
    "stated_constraints": [
      "energy_target_range",
      "sodium_limit",
      "vegetarian_preference"
    ],
    "missing_information": [
      "current_medication_list",
      "recent_lab_values"
    ]
  },
  "instruction": "Produce a structured plan outline that respects the stated constraints and lists information required before finalising. System prompts and hidden generation instructions are excluded from released records.",
  "candidate_response": {
    "required_clarifications": [
      "current_medication_list",
      "recent_lab_values"
    ],
    "plan_outline": "<structured_outline>",
    "constraints_applied": [
      "energy_target_range",
      "sodium_limit",
      "vegetarian_preference"
    ]
  },
  "automated_checks": {
    "schema_valid": true,
    "constraints_referenced": true,
    "missing_info_acknowledged": true,
    "numeric_bounds_ok": true
  },
  "review_status": "automated_only",
  "generation_version": "gen-v3"
}

Generation methodology

1. 01Scenario specification defines scope, constraints, difficulty, and intended task.
2. 02Controlled synthetic generation produces fictional scenarios and candidate responses.
3. 03Automated schema and completeness checks run on every record.
4. 04Deterministic numerical constraints are verified where the task defines them.

Validation information

• Internal rubric-based evaluation on sampled subsets during R&D.
• Qualified dietitian review available for scoped engagements; coverage agreed per dataset.
• No claim of clinician review across the full corpus.

Known limitations

• Synthetic generation does not establish factual medical correctness.
• Guideline coverage reflects the focus areas explored internally, not comprehensive clinical coverage.
• Spanish-language records have narrower automated check coverage than English records.
• Not suitable as a sole basis for clinical decision support.

Licensing and access process

Licence status: Research access by inquiry. Access is sales-assisted. Licensing is handled by agreement, based on data provenance, rights, permitted uses, delivery, and commercial terms.

1. Step 1Submit an inquiry describing the model, task, and intended use.
2. Step 2Scoping call to confirm fit, formats, and review requirements.
3. Step 3Rights and provenance review before any sample or delivery.
4. Step 4Access and licence terms agreed in writing.