A response driven only by a lagged measurement moves the wrong way at the start of a step

Fact

When a system's response is computed from a measurement of its own input, and that measurement lags the input, the response lags it too — by the sum of both time constants. If the quantity being regulated depends on the ratio of the two, it moves in the wrong direction until the slower signal catches up.

A ventilation model driven from oxygen uptake showed it clearly: uptake rises with a ~30 s time constant, ventilation was given a 12 s lag on top of it, so at the onset of hard work CO2 production outran ventilation and arterial CO2 rose by 27 mmHg while saturation fell to 88% — the opposite of what starting to exercise does.

Why it matters

Both lags are individually correct and defensible, and the steady state is right, so nothing looks wrong except the transient. The symptom lands in the regulated variable, several steps downstream, so it gets attributed to that part: to the gas-exchange equations, to a threshold, to a limit somewhere.

How to apply

• Feed the response forward from the demand, not from the measured consequence of it: drive = max(measured, demanded). Real regulators do the same thing — the neural component of exercise hyperpnoea exists precisely because the chemical one is too slow.
• Lag the feed-forward term too, but less than the process it anticipates, or the response steps to its final value while the load is still at rest and overshoots as hard in the other direction.
• Close the loop on the regulated variable itself, so any residual error is corrected rather than integrated.
• Test the transient, not only the steady state: step the input and assert the regulated variable stays in range throughout.