A curve can fit and still miss the pattern
Fit a cooling curve, inspect its errors, then see how it fares on points it never saw.
Synthetic cooling experiment
Can the fit predict the points it hasn’t seen?
Choose how the sample cools. The orange line always fits one cooling rate. Move the last training time to change which points it can use.
What the model assumes
Ambient temperature is fixed at 20 °C. The known initial temperature is 80 °C. The fit chooses a single rate to minimize squared temperature errors on the training points. A small, fixed perturbation gives each example the same repeatable noise. These are synthetic values. Neither small errors nor a good prediction establishes the unique physical mechanism.
What does a positive residual tell you?
Taking a real process
beyond the lab?
Jon and I are building RARE Labs toward an autonomous pilot plant builder using AI and robotics. Join the waitlist for updates, and tell us what you’re trying to scale.