Apparatus build-and-defense
This is a live exam at the bench. The student designs and builds a rig to measure a physical quantity — a pendulum to pin down g, or a cart-and-track to measure the friction between two surfaces — then takes real data and defends it. The guide starts asking: why that length and that release angle, why time many swings instead of one, how the result falls out of the numbers, and how far off it could be. There is no worksheet to copy and no figure to look up: the student stands over the apparatus they built and defends the run out loud.
| Criterion | Not yet | Approaching | Mastered |
|---|---|---|---|
| Apparatus design | The rig does not actually isolate the quantity being measured, or confounds are left uncontrolled. | The design isolates the variable but a parameter is poorly chosen, or one confound is left unaddressed. | The rig cleanly isolates the variable, chooses sensible parameters, and controls the confounds that would otherwise creep in. |
| Data collection | Takes a single reading, uses technique that magnifies error, and keeps obvious outliers. | Repeats trials but inconsistently, or uses a partial error-shrinking technique — times a few periods rather than many. | Runs repeated trials, uses technique that shrinks error — timing many periods at once — and handles outliers deliberately. |
| Calculation & uncertainty | Reaches a wrong result, reports no uncertainty, or writes digits the data cannot support. | Reaches a plausible result but omits the uncertainty or carries too many significant figures. | Derives the correct result, propagates the uncertainty through the calculation, and reports it to the right number of significant figures. |
| Oral defense under questioning | Folds at the first follow-up or recites a memorized line that does not fit the rig they built. | Answers some follow-ups, falters when asked to justify a design choice or handle a “what if.” | Explains and justifies each choice under questioning and reasons through unrehearsed “what if” probes on the spot. |
“I timed twenty full swings and divided, because one swing is too fast to time by hand — that alone cuts my timing error by twenty. From the period and the length, g comes out to 9.79 meters per second squared, and my uncertainty is about two percent, mostly from measuring the length to the center of the bob.”
“I let it swing and timed it once. I got something around 9-point-something. I’m not really sure how off it is or where the error comes from.”
This assessment is AI-proof by design: it happens at the bench, with a rig the student built and real data they took, in real time. No chatbot can clamp a pendulum, justify a number it did not measure, or hold up under a follow-up question about a design choice it did not make. Mastery is shown by building and defending — not by submitting.