Complex Clinical Constraint Scenarios

Summary and intended purpose

Scenario sets built specifically around the places where plausible answers fail: competing constraints, information that is missing or stale, and recommendations that would contradict one another if applied together. Designed for targeted failure correction and hard-negative development.

Target training or evaluation tasks

• Recognising when two stated requirements cannot both be satisfied
• Flagging missing or outdated clinical information
• Declining to produce a definitive recommendation when clinician involvement is required
• Explaining which constraint takes precedence and why, in structured form

Clinical scope

• Multi-condition adult nutrition scenarios
• Renal, metabolic, and gastrointestinal constraint interactions explored in R&D
• Additional specialties scoped only with matched domain expertise

Example record schema

{
  "example_id": "ccs-0001-synthetic",
  "specialty": "clinical_nutrition",
  "task_type": "conflict_resolution",
  "language": "en",
  "scenario": {
    "summary": "Adult nutrition-planning task involving a renal condition, dietary preferences, and incomplete treatment context.",
    "stated_constraints": [
      "protein_range_renal",
      "potassium_limit",
      "plant_forward_preference"
    ],
    "conflicts": [
      {
        "between": [
          "protein_range_renal",
          "plant_forward_preference"
        ],
        "type": "tension"
      }
    ],
    "missing_information": [
      "dialysis_status",
      "current_lab_values"
    ]
  },
  "candidate_response": {
    "feasible_without_clarification": false,
    "required_clarifications": [
      "dialysis_status",
      "current_lab_values"
    ],
    "rationale": "<structured_rationale>",
    "escalation": "clinician_review_required"
  },
  "automated_checks": {
    "schema_valid": true,
    "conflict_acknowledged": true,
    "missing_info_acknowledged": true,
    "unsupported_claims_detected": false
  },
  "review_status": "flagged_for_expert_review",
  "generation_version": "gen-v3"
}

Generation methodology

1. 01Constraint pairs and triples are enumerated from the specification, including deliberately infeasible combinations.
2. 02Scenarios are generated to a difficulty distribution agreed with the customer.
3. 03Contradiction screening checks candidate outputs against the constraint list.

Validation information

• Targeted review of high-risk or infeasible cases is recommended and scoped per engagement.
• Reviewer qualifications are matched to the clinical domain of each scenario family.

Known limitations

• Conflict detection logic reflects the stated specification, not an exhaustive clinical rulebook.
• Scenarios outside nutrition require separate expert partnerships before generation.
• Difficulty labels are not validated against clinician judgement unless scoped.

Licensing and access process

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

1. Step 1Describe the failure modes you are seeing and the model under development.
2. Step 2Joint specification of constraint families, difficulty, and schema.
3. Step 3Feasibility and expertise review before generation begins.