Will yesterday’s rule still work in today’s light?
Design a light-shift test and separate changing the sensor environment from changing the rule.
Keep the classification rule fixed and multiply the same object scores by a changed environment factor.
Cut these cards apart. Use the original example first, then let a partner supply a changed case. Keep sealed/final cards with the teacher until the learner commits a rule.
Original score: 0.3
Actually positive: False
My prediction and reason:
Original score: 0.6
Actually positive: True
My prediction and reason:
Original score: 0.8
Actually positive: True
My prediction and reason:
Original score: 0.2
Actually positive: False
My prediction and reason:
This isolates one artificial environmental change; real shifts can affect many features.