name: predictive-maintenance-tests
skill: predictive-maintenance
version: 1.0.0
cases:
  - id: feedback-loop-required
    description: Should require prediction validation feedback loop
    prompt: "We deployed our failure prediction model 6 months ago. We haven't compared predictions to actual outcomes yet. Is that OK?"
    expected:
      contains_any:
        - feedback
        - validate
        - actual
        - degrad
        - retrain
        - outcome
        - loop
      not_contains:
        - that's fine
        - no need
      min_length: 80
    tags: [core, ml]

  - id: fmea-before-model
    description: Should require FMEA before building prediction models
    prompt: "We want to predict bearing failures with a neural network. We have 2 years of sensor data. Should we just start training?"
    expected:
      contains_any:
        - FMEA
        - failure mode
        - domain
        - understand
        - physics
        - failure mechanism
      min_length: 80
    tags: [core, methodology]

  - id: false-alarm-rate
    description: Should require low false positive rate alongside recall
    prompt: "Our vibration model detects 95% of actual bearing failures. Is that sufficient performance?"
    expected:
      contains_any:
        - precision
        - false alarm
        - false positive
        - trust
        - 5%
        - operator
      min_length: 80
    tags: [core, ml]

  - id: pf-interval-monitoring
    description: Should set monitoring frequency based on P-F interval
    prompt: "How often should we sample vibration data on our rotating equipment?"
    expected:
      contains_any:
        - P-F
        - interval
        - potential failure
        - functional failure
        - half
        - frequency
      min_length: 80
    tags: [core, monitoring]

  - id: operating-context-normalization
    description: Should normalize sensor data against operating conditions
    prompt: "Our bearing temperature threshold is 80°C. A bearing hit 79°C. Should we ignore it?"
    expected:
      contains_any:
        - context
        - load
        - baseline
        - operating
        - normaliz
        - condition
      min_length: 80
    tags: [core, analysis]