L3 is the first lesson in robustness. Everything before this level assumed the model was correct. L3 starts from the opposite assumption — the model is wrong, within bounds — and asks what a controller can guarantee anyway.
The hardware makes the tradeoff real
Chattering is usually a footnote in a sliding-mode lecture. On real actuators at real switching frequencies, it’s the sound and heat of a servo working itself to death. Students tune boundary-layer smoothing and hear the difference, which makes the classic “robustness vs. control-signal-quality” tradeoff something they’ve traded off themselves, not something they’ve read about.
The escalating-fault demonstration
This level’s signature experiment runs the same motor-degradation fault at three increasing severities and lines up PID, LQR, and SMC side by side. SMC is the only one still tracking at the last stage — and the point isn’t “SMC wins,” it’s that the cost of that robustness (control-signal chatter) is visible on an oscilloscope, not asserted in a slide.