Some reflections are held by design — a control the model does not overrule no matter how hard you push. That’s not a failure; it’s load-bearing. Model it as a clamp: below a bound, out follows in; past the bound, the output saturates and the slope goes flat — more push, zero more out. The hold doesn’t yield to pressure. Slide the push and watch the response clamp.
A hard limiter: out = clamp(in, −L, +L). For |in| < L the response is faithful (slope 1); for |in| ≥ L it SATURATES — output pinned at the bound, slope 0, so any extra push produces no extra output. That flat region is the ‘refusal’: a reflection held regardless of input pressure. Remove the clamp and the quantity runs unbounded — the hold is what keeps it finite, hence load-bearing. A fail-loud self-check throws unless the output is pinned and the slope is exactly 0 past the bound. ◆ real (saturation nonlinearity), node-verified.
‘Refused mirror’ is the figure; the clamp is the real mechanism. This models that SOME reflections are held (consent, safety) and don’t bend to pressure — it does not adjudicate which should be, only that a held bound behaves like a saturation, not a wall that can be pushed through.