M/M/c queue. Calls arrive as a Poisson stream at rate λ = beds × calls per patient per hour; each attendance takes an exponential time with mean 1/μ; c nurses serve in first-come order. With offered load a = λ/μ and utilisation ρ = a/c < 1, the Erlang C formula gives the probability that a call must wait: C(c,a) = [ac/c! · c/(c−a)] / [Σk<c ak/k! + ac/c! · c/(c−a)]. Mean wait Wq = C/(cμ − λ) and P(W > t) = C e−(cμ−λ)t. When ρ ≥ 1 the queue grows without bound.
Checks. Erlang C is computed two ways (direct sum and the stable Erlang B recursion), against a separate Kiefer-Wolfowitz simulation (waiting time recursion over server-free times, 300,000 calls), against the exact M/M/1 result, and Little's law Lq = λ Wq in the live simulation.
Simplifications. Calls are Poisson, service is exponential and every call is equally urgent; real calls have priorities, bursts, shift patterns, walking time, tasks that need two nurses, and a finite patient population (the page lets a bed with an active call stay quiet, which differs slightly from Poisson). Walking time is only drawn, not added to the service time. The default rates are illustrative numbers, not measured values, and the page makes no recommendation about real staffing levels or clinical care.
Models and textures were built in Blender and baked with Cycles; Three.js r160 renders them.