skills/jinko-model/evals/evals.json

{
  "skill_name": "jinko-model",
  "evals": [
    {
      "id": 1,
      "prompt": "I have an existing Jinkō model cm-123 and want to retune k_clearance plus add k_new. Show me the safest SDK way and avoid one request per edit.",
      "expected_output": "Uses model.components.batch with edit_parameter and create_parameter, requires --apply or explicit confirmation for mutation, and avoids raw payloads."
    },
    {
      "id": 2,
      "prompt": "Create a tiny new SDK model with a dose event, ODE, patient-varying elimination, unit checking, and a solvability check. Tag the protocol input, the patient input, their value sources, and the output. I do not want biological guidance.",
      "expected_output": "Routes to create_empty_model, uses UnitCheckAndConvertAllSpeciesToExtentUnits, gives numeric component values units, attaches the built-in i::protocol, i::vpop, s::arbitrary, and output tags without recreating them, creates components in a batch, then checks diagnostics and simple_solve."
    },
    {
      "id": 3,
      "prompt": "Jinkō reports TOO_FEW_ALGEBRAIC_RULES and MISSING_UNKNOWN. How should I encode algebraic rules in the SDK?",
      "expected_output": "Explains residual-equation form 0 = equation, one algebraic rule per unknown variable, and suggests inspecting diagnostics before adding or deleting rules."
    },
    {
      "id": 4,
      "prompt": "Add a breakfast event at 7 h that switches the meal_rate parameter from 0 to 1 mol/h, then prove it happened in the model solve.",
      "expected_output": "Creates a mutable parameter with constant=False, uses a dimensioned time trigger, record=True, and validates the returned meal_rate series before and after the event."
    },
    {
      "id": 5,
      "prompt": "My model uses a 0.06 per minute absorption rate. How should I add it safely and check it is not interpreted per second?",
      "expected_output": "Selects intentional unit checking, uses an inverse-time unit, and requires diagnostics plus a representative numerical solve rather than trusting the unit label."
    },
    {
      "id": 6,
      "prompt": "Tag Dose as i::protocol and s::knowledge, CL as i::vpop and s::to-calibrate, and Drug as output in cm-123. Preserve unrelated tags and do not create a separate edit per component.",
      "expected_output": "Reuses the built-in standard model tags, uses one component batch to replace conflicting scoped tags while preserving unrelated tags, and does not recreate platform tag declarations."
    },
    {
      "id": 7,
      "prompt": "Turn off unit checking for my new model because I have not decided units yet.",
      "expected_output": "Does not silently select NoUnitCheck. Explains the default UnitCheckAndConvertAllSpeciesToExtentUnits policy and waits for an explicit human decision before choosing another mode."
    },
    {
      "id": 8,
      "prompt": "Run the model readiness script without any --require-tag options and tell me whether the model is fully ready.",
      "expected_output": "Checks diagnostics and simple_solve but states that the tag policy was not checked, rather than claiming full model readiness."
    }
  ]
}