---

Claim. For every $C>0$ there are $\epsilon>0$ and $n_0$ such that for every $n\ge n_0$ and every choice of nodes $x_1,\ldots,x_n\in[-1,1]$, counted with multiplicity, there are labels $y_1,\ldots,y_n\in[-1,1]$ such that every real or complex polynomial $P$ with $\deg P<(1+\epsilon)n$ and $P(x_i)=y_i$ for at least $(1-\epsilon)n$ indices has $\max_{[-1,1]}\lvert P\rvert>C$. This is Theorem 1.1 of the note A Bernstein-density proof of Erdős's robust interpolation obstruction (29 April 2026, no author byline, hosted at ulam.ai), the assertion of Problem 1133 with repeated nodes allowed. The card anon_2026_bernstein_density_proof_erdos_s_robust digests the argument. Beurling's theorem, in the real-line form of Ortega-Cerdà and Seip, says that a separated sequence interpolates the Bernstein space $B_1$ only when its upper uniform density is below $1/\pi$; by compactness this yields a finite obstruction (Proposition 3.1): for each $C$ there are $L$ and $\eta>0$ such that any $L$ reals (with multiplicity) of diameter at most $\pi(1+\eta)L$ receive labels in $[-1,1]$ that no complex-valued $f\in B_1$ of sup norm at most $C$ takes at all of them. After $x=\cos\theta$ the nodes are cut into consecutive blocks of $L$; more than $\epsilon n$ blocks have angular span at most $\pi(1+\eta)L/\lceil(1+\epsilon)n\rceil$, each is loaded with a forbidden pattern, and after angular rescaling a polynomial of norm at most $C$ fitting a whole good block would give a function in $B_1$ of norm at most $C$ taking the forbidden labels. The note gives no quantitative dependence of $\epsilon$ on $C$; the later manuscript of Yang, which cites it as an unsigned draft and reuses its angular reduction and grouping, supplies that dependence.

Depends on. No page of this wiki: the one external input is Beurling's interpolation theorem, cited from the literature.

Claimant. Przemek Chojecki, who posted the note in the site's thread on 29 April 2026 and wrote that GPT-5.5 Pro produced the argument through Beurling's interpolation theorem for Bernstein spaces; the note carries no byline and its hosting page names no author.

Standing. A manuscript claim, claimed. A reply in the thread the same day reports that a tool-assisted check, a linked chat transcript, found no issues in the note; that is not a review. The card also records a listing of the note as an AI-assisted candidate solution on a public wiki of AI contributions, which is not a review either. Not reviewed: the site's label is OPEN (page last edited 31 December 2025), its commentary does not mention the note, and the note is not on the site's proof-claims tab, which carries only Yang's later claim. Not refereed: no journal or arXiv version is known. The proof has not been reviewed.