Customers enter one at a time. Each joins an occupied table with weight (its size − d) or opens a new table with weight α + d K, where K is the number of tables so far. With discount d = 0 this is the Chinese restaurant process of a Dirichlet process, and the number of tables grows like α log N; with d > 0 (Pitman-Yor) it grows like Nd and the table sizes follow a power law. The solid curve is the exact expectation EKn; the jagged one is your restaurant.
The expectation uses the exact recursion EKn+1 = EKn (1 + d/(α + n)) + α/(α + n). For d = 0 the standard deviation is the square root of the sum of α i/(α + i)², because the new-table events are independent.
code/ch31_nonparametric_bayes.py (crp_mulberry, expected_tables, Figure x31.1).