Experiment-design defense
This is a live exam in the lab. The student brings a question of their own — does raising the ramp make the toy car roll farther? — and designs a fair test for it: one variable changed, the rest held steady, enough trials to trust the answer. They run it, record the data, and graph it. Then the guide starts asking: why hold that constant, why five trials and not one, what does that outlier mean, what if the floor were carpet instead. There is no worksheet to copy and no answer to look up — the student stands over their own setup and defends every choice out loud.
| Criterion | Not yet | Approaching | Mastered |
|---|---|---|---|
| Testable question & hypothesis | Starts with a question no experiment could answer, or offers no hypothesis at all. | Has a real question but leaves it too broad, or predicts an outcome without an if/then. | States a sharp, testable question and an if/then hypothesis that an experiment could prove wrong. |
| Sound design (variables, controls, replication) | Changes several things at once, or cannot say what is being held constant. | Identifies the main variable but misses a control, or runs only a single trial. | Changes one variable, holds the rest steady, and repeats the run enough times to trust the result. |
| Clean data & appropriate graph | Records data carelessly, or picks a graph that hides the pattern. | Records the data but the table is disorganized, or the graph is mislabeled. | Records tidy, labeled data and graphs it with the right chart type, so the pattern is easy to read. |
| Justified conclusion | States a conclusion the data does not support, or draws none at all. | Draws a reasonable conclusion but does not point to the data that backs it. | Draws a conclusion that follows from the data, names the evidence for it, and says what the result does not prove. |
| Live oral defense | Folds at the first “what if,” or recites a line that does not fit the run. | Answers some questions but falters when asked to justify a design choice. | Fields unrehearsed “what if” questions about this experiment with sound, on-the-spot reasoning. |
“My question was whether a steeper ramp makes the car roll farther. I only changed the ramp height — same car, same floor, same starting push — and I ran each height five times so one weird roll wouldn’t fool me. The data says higher ramp, farther roll, every time. If the floor were carpet I’d expect shorter distances, but the pattern would probably still hold.”
“I made the ramp higher and the car went farther, I think. I only did it once. I’m not really sure what I kept the same.”
This assessment is AI-proof by design: it happens in the lab, with a real setup the student built, in real time. No chatbot can hold a variable steady, run the trials, or defend a design choice under a “what if” it cannot picture. Mastery is shown by designing, running, and defending — not by submitting.