A visual companion to the "catenary as a neural-network activation function" idea: two draggable points A and B define the endpoints of a hanging-chain (catenary) curve. The curve is clamped flat outside [x1, x2] and blended into those plateaus with a cubic Hermite ease, so the whole thing behaves like a smooth, sag-tunable activation function rather than an open-ended arc.

Controls: drag move point A or B (the curve's endpoints) scroll (away from a point) adjust the chain length / sag amount scroll (over a point) resizes that point's handle, ignore it

The catenary math (solveCatenary / catenaryOutput / catenaryDerivative) and the Hermite tip-smoothing formulas are ported near-verbatim from the catenary.ts activation-function experiment - see the session notes for the closed-form derivation. Canvas y grows downward, so points are solved in a y-flipped "math space" and flipped back for drawing - this makes the chain sag down on screen like a real rope, rather than arching upward.