Read and reason
Control which dataset you study
The studio includes two synthetic motion examples. One follows an approximately uniform-velocity trend and one an approximately constant-acceleration trend, both with small invented deviations. The tool plots position against time, fits linear and quadratic curves using five training points, and checks the prediction at a sixth held-out point. You may inspect both datasets to see that the curve with more coefficients is not automatically the best scientific choice.
Read a residual plot
Plot residual r versus time. A residual close to zero indicates that prediction was close at that moment; a pattern of alternating positive and negative values may indicate overfitting or limitations, depending on context. The residual graph's vertical axis is measured in meters. You should discuss whether its scale is reasonable compared with measurement resolution and the time interval rather than looking only at how smooth the model curve appears.
Make a defensible selection
Report your chosen model, training RMSE, absolute holdout error, number of fitted parameters, and plausible physics. For the intentionally near-linear dataset, favor parsimony unless evidence supports acceleration. For the curved dataset, a quadratic mechanism may be reasonable if its prediction and residual pattern improve meaningfully. A six-point invented dataset cannot prove a general physical law.
Safety, transparency and reusability
The model uses data generated from formulas, not actual video or motion-sensor readings. Its mathematical process is reproducible and standalone. Never collect identifiable video of others without permission; recording real motion requires consent and suitable supervision. The dataset CSV is available offline so a student can independently calculate sample residuals without running the interactive graph.