Before increasing gain, write down what the sensor reports. The controller sees the measured signal; a number called “error” may not mean the physical target minus the output.
One change that exposes the distinction
Open Block algebra, choose Negative feedback, and keep G₁=2 and G₂=3. At H=1, y=6/7 and both errors equal 1/7. Record a prediction that output will decrease, then set H=2 and check it. The output becomes 6/13≈0.462. The summing error is 1/13≈0.077, while physical tracking error is 7/13≈0.538.
The loop is doing what it was wired to do
The summing point computes e=r−Hy. Substituting y=G₁G₂e gives y=G₁G₂r/(1+G₁G₂H). A smaller e does not imply smaller r−y unless the sensor and reference share the intended scale. In a speed loop this can happen when rpm, rad/s and a sensor voltage are mixed without explicit calibration.
A practical debugging order
Write the units at every wire. Check the sensor conversion, the reference scale and the summing sign before changing gains. Reproduce the same input after calibration. This bench uses static gains, so it isolates scaling; dynamic stability still requires the plant and controller poles.
Compare r−y and r−Hy before declaring tracking success.