The site's wording puts no normalization on $p$. The change writes $p(z)=z^n+\ldots$, so that $p$ is monic; nothing else changes. The evidence is Erdős's own statement of the problem. [Er61], Problem IV.1, p. 246 (Some unsolved problems), opens "Let $z^n+\ldots$ be a polynomial of degree $n$" and asks on p. 247 whether $\max_{z\in E_f^{(n)}}\lvert f'(z)\rvert<\frac{n^2}{2}$ when $E_f^{(n)}$ is connected, his (IV.1.1). Hayman's collection states the question as Problem 4.8, one of five problems on the set $E_f^{(n)}$, the first of which, Problem 4.7, takes $f(z)=z^n+a_1z^{n-1}+\ldots+a_n$ (Research Problems in Function Theory, the 2018 edition, which keeps the 1967 numbering), and Eremenko and Lempert state it from there with $f(z)\sim z^n$ as $z\to\infty$, their (1) on p. 191. The site's own commentary fits only monic polynomials: it says the maximum is at least $n$, with equality only for $p(z)=z^n$, and for $cz^n$ with $\lvert c\rvert<1$ the maximum is $n\lvert c\rvert^{1/n}<n$. The defect is the site's; Erdős's text carries the normalization. The form comes from [Er61], not from the hypotheses of the theorem that settles it. No result about the site's wording is recorded.