\documentclass[11pt]{article} \usepackage[T1]{fontenc} \usepackage[utf8]{inputenc} \usepackage{lmodern} \usepackage[margin=1in]{geometry} \usepackage{amsmath,amssymb,amsthm,mathtools} \usepackage{enumitem} \usepackage{booktabs} \usepackage{array} \usepackage{fancyvrb} \usepackage{float} \usepackage{xcolor} \usepackage{xurl} \usepackage{microtype} \usepackage{fancyhdr} \usepackage[colorlinks=true,urlcolor=blue!50!black,linkcolor=blue!50!black,citecolor=blue!50!black]{hyperref} \hypersetup{pdftitle={Sharp finite cases of the Erdos-Suranyi interval-product problem: g(4)=5 and g(5)=6}, pdfauthor={John Tripodo}, pdfsubject={Erdos Problem 708: exact values g(4)=5 and g(5)=6, with Lean 4 verification}, pdfkeywords={Erdos Problem 708, Erdos-Suranyi function, interval product-cover, Hall's theorem, Lean 4, Palomar}} % Unicode characters that occur in the verbatim Lean listing (pdfLaTeX). \DeclareUnicodeCharacter{2115}{\ensuremath{\mathbb{N}}} \DeclareUnicodeCharacter{2124}{\ensuremath{\mathbb{Z}}} \DeclareUnicodeCharacter{2200}{\ensuremath{\forall}} \DeclareUnicodeCharacter{2203}{\ensuremath{\exists}} \DeclareUnicodeCharacter{2208}{\ensuremath{\in}} \DeclareUnicodeCharacter{220F}{\ensuremath{\prod}} \DeclareUnicodeCharacter{2223}{\ensuremath{\mid}} \DeclareUnicodeCharacter{2264}{\ensuremath{\le}} \DeclareUnicodeCharacter{2286}{\ensuremath{\subseteq}} \DeclareUnicodeCharacter{2192}{\ensuremath{\to}} \DeclareUnicodeCharacter{2227}{\ensuremath{\wedge}} \newtheorem{theorem}{Theorem}[section] \newtheorem{lemma}[theorem]{Lemma} \newtheorem{proposition}[theorem]{Proposition} \newtheorem{corollary}[theorem]{Corollary} \theoremstyle{definition} \newtheorem{definition}[theorem]{Definition} \newtheorem{example}[theorem]{Example} \theoremstyle{remark} \newtheorem{remark}[theorem]{Remark} \DeclareMathOperator{\lcm}{lcm} \newcommand{\Z}{\mathbb{Z}} \newcommand{\repo}{\url{https://github.com/jjtripodo-lgtm/erdos-708-g4-g5}} \newcommand{\commit}{00fbb23c3084daa88ea420aa9c74177127d5283f} \newcommand{\palomarid}{\mbox{PALOMAR-2026-11-19-000019}} \title{Sharp finite cases of the Erd\H{o}s--Sur\'anyi\\ interval-product problem:\\ $g(4)=5$ and $g(5)=6$} \author{John Tripodo\thanks{ORCID: 0009-0006-9180-4953. \textcopyright\ 2026 John Tripodo. This manuscript is distributed under the Creative Commons Attribution 4.0 International licence (CC BY 4.0). The accompanying Lean source is licensed under Apache-2.0.}} \date{19 November 2026\\\small Manuscript version 1.8} \pagestyle{fancy} \fancyhf{} \fancyhead[L]{\small Erd\H{o}s--Sur\'anyi finite cases} \fancyhead[R]{\small Preprint, version 1.8} \fancyfoot[C]{\thepage} \setlength{\headheight}{14pt} \begin{document} \maketitle \begin{abstract} Let $g(n)$ be the Erd\H{o}s--Sur\'anyi function of Erd\H{o}s Problem~\#708: the least $k$ such that, for every set $A$ of $n$ distinct integers at least $2$ and every interval $I$ of $\max A$ consecutive integers, some set of at most $k$ elements of $I$ has product divisible by $\prod_{a\in A}a$. We prove \[ g(4)=5\qquad\text{and}\qquad g(5)=6. \] The lower bound $g(n)\ge n+1$ for $n\ge3$ is due to Erd\H{o}s and Sur\'anyi (1959). Our contribution is the matching universal upper bounds $g(4)\le5$ and $g(5)\le6$; to our knowledge these are the first exact values beyond $g(3)=4$. The four-modulus proof combines a Hall-matching reduction with gcd/lcm collision repair and an anchored proper-divisor lemma. The five-modulus proof classifies a maximal antichain-admissible assignment of the moduli to interval multiples by its occupancy shape; the hardest shapes are a squarefree four-prime exception and a $(4,1)$ configuration controlled by a propagation invariant. Both values, in the exact-cardinality reading of the original problem and in the at-most reading used by later work, are machine-checked in Lean~4. The formalization is registered in the Palomar registry as \palomarid. The general question whether $g(n)\le 2n$ remains open. \end{abstract} \noindent\textbf{Keywords:} Erd\H{o}s Problem 708; Erd\H{o}s--Sur\'anyi function; interval product-cover; Hall's theorem; divisibility antichains; Lean~4; formal verification. \noindent\textbf{MSC 2020:} 11A05 (primary); 11B75, 05D15 (secondary). \medskip \noindent\textbf{Status of the results.} Theorem~\ref{thm:main} is formally verified: four Lean declarations stating it were mechanically checked at a pinned public commit, and the formalization is registered as \palomarid\ (Section~\ref{sec:formal}). Sections~\ref{sec:four}--\ref{sec:sharp5} give a human-readable account of the same mathematics. Some auxiliary lemmas are organized differently from their Lean counterparts (Table~\ref{tab:map}), and the written text has not been refereed. The provenance of the work, including the use of AI systems, is disclosed at the end of the paper. \section{The problem and two cardinality conventions}\label{sec:problem} For an integer $L$ and a positive integer $M$, write \[ I(L,M)=\{L,L+1,\ldots,L+M-1\}. \] For a finite nonempty set $A\subseteq\{2,3,\ldots\}$, put $M=\max A$ and $P(A)=\prod_{a\in A}a$. An \emph{interval product-cover} of $A$ in $I(L,M)$ is a subset $B\subseteq I(L,M)$ with \[ P(A)\mid\prod_{b\in B}b. \] All products are integer products. Intervals may start at any integer; if $0\in I(L,M)$, then $\{0\}$ is a product-cover. Erd\H{o}s's 1992 formulation \cite{Er1992} uses positive starts. Reflecting an interval through $0$ changes products only by a sign, so the two conventions give the same values. \begin{definition}[At-most reading]\label{def:g} Let $C(n,k)$ be the statement: for every $n$-element $A\subseteq\{2,3,\ldots\}$ and every integer $L$ there is a product-cover $B\subseteq I(L,\max A)$ with $|B|\le k$. Put \[ g(n)=\min\{k:C(n,k)\}. \] Here $n\ge1$, and the set of admissible $k$ is nonempty; for example, Chen proves $g(n)\le12n$ \cite{Chen708}. \end{definition} \begin{definition}[Exact reading]\label{def:gexact} Let $C^{=}(n,k)$ be the same statement with $|B|=k$ in place of $|B|\le k$. When some $k$ satisfies $C^{=}(n,k)$, let $g^{=}(n)$ be the least such $k$. \end{definition} The problem as printed in the tracker uses the exact condition \cite{Bloom708}. Later work uses the at-most condition \cite{Chen708,FC4388,White708}. The two differ in general. For $A=\{2,3,\ldots,n+1\}$ every relevant interval has only $n+1$ elements, so $C^{=}(n,k)$ fails for every $k>n+1$. Hence, whenever $g(n)>n+1$, no $k$ satisfies $C^{=}(n,k)$ and $g^{=}(n)$ is undefined. By the asymptotic lower bound $g(n)\ge(2-o(1))n$ of Erd\H{o}s and Sur\'anyi \cite{ErSu1959}, this happens for all large $n$. \begin{lemma}[Padding]\label{lem:padding} Let $n\ge1$ and $k\le n+1$. Then $C(n,k)$ holds if and only if $C^{=}(n,k)$ holds. \end{lemma} \begin{proof} An exact cover of size $k$ is a cover of size at most $k$. Conversely, $n$ distinct integers that are at least $2$ have maximum $M\ge n+1\ge k$, so $I(L,M)$ has at least $k$ elements. A cover $B$ with $|B|M$, the two candidate families are disjoint, and each has two elements. \end{itemize} Hence the union always has at least $t$ elements, and Hall's theorem gives distinct representatives. \end{proof} \begin{lemma}[Common gcds in the upper half]\label{lem:upperhalf} Two distinct integers in $(M/2,M]$ have gcd at most $M/3$. Three distinct integers in this range have gcd at most $M/5$. \end{lemma} \begin{proof} After division by the common gcd $d$, the quotients are distinct integers in $(M/(2d),M/d]$. If $M/d<3$ this range contains at most one integer, and if $M/d<5$ it contains at most two. \end{proof} \begin{lemma}[Four moduli at one anchor]\label{lem:oneanchor} If four distinct moduli have the same unique anchor $x$, then five sites suffice. \end{lemma} \begin{proof} All four moduli exceed $M/2$, so they form a divisibility antichain. Let $Q=\lcm(a_1,\ldots,a_4)$, so $Q\mid x$. Connect $a_i$ to the prime units $(p,j)$ of $Q$ with $j\le v_p(a_i)$ (prime units are defined before Lemma~\ref{lem:fourunits}). By Lemma~\ref{lem:fourunits}, Hall's theorem assigns distinct units to the four moduli. Let $p_i$ be the prime of the unit assigned to $a_i$; then $\prod_ip_i\mid Q$. Put $r_i=a_i/p_i$. By Lemma~\ref{lem:proper}, there are four distinct multiples of the proper divisors $r_i$ off $x$. Their product pays $\prod_ir_i$, and $x$ pays $\prod_ip_i$. Together the five sites pay $a_1a_2a_3a_4$. \end{proof} \begin{lemma}[Triple-anchor assembly]\label{lem:triple} Suppose three of the moduli, $a,b,c$, have the same unique anchor $x$, and the fourth has an anchor $y\ne x$. Then five sites suffice. \end{lemma} \begin{proof} Write $h=\gcd(a,b,c)$ and write the three pairwise gcds as $hu,hv,hw$. Comparing prime valuations shows that $u,v,w$ are pairwise coprime and \begin{equation}\label{eq:tripleid4} abc=\lcm(a,b,c)\,h^2uvw. \end{equation} Each pairwise gcd is at most $M/3$ by Lemma~\ref{lem:upperhalf}, so it has a multiple outside $\{x,y\}$; choose one for each. Three distinct choices pay $h^3uvw$. If exactly two sites occur, the shared site pays the lcm of two of the labels and the other site pays the third, which gives $h^2uvw$. If all three choices coincide, that site pays $\lcm(hu,hv,hw)=huvw$, and since $h\le M/5$ there is another multiple of $h$ outside the three occupied sites. In every case at most three sites outside $\{x,y\}$ pay $h^2uvw$. The anchor $x$ pays $\lcm(a,b,c)$ and $y$ pays the fourth modulus, so by \eqref{eq:tripleid4} at most five sites suffice. \end{proof} \begin{lemma}[Two-anchor completion]\label{lem:twoanchor} Suppose the union of the candidate sets of the four moduli is exactly $\{x,y\}$, and no three moduli have a common unique anchor. Then five sites suffice. \end{lemma} \begin{proof} Put $D=|y-x|$. A modulus dividing both $x$ and $y$ whose candidate set is $\{x,y\}$ must equal $D$; otherwise a multiple strictly between $x$ and $y$ would lie in $I$. Call such a modulus a \emph{bridge}. Since the moduli are distinct, there is at most one bridge. \emph{No bridge.} Then every modulus has a unique anchor, and since no three share one, two moduli $a,b$ are anchored at $x$ and two moduli $c,d$ at $y$. The gcds $r=\gcd(a,b)$ and $s=\gcd(c,d)$ are at most $M/3$, so each has a multiple off $\{x,y\}$. If the chosen sites differ, they pay $rs$, and with $x$ and $y$ (paying $\lcm(a,b)$ and $\lcm(c,d)$) this covers $abcd$. If they coincide, that site pays $\lcm(r,s)$. The gcd $q=\gcd(r,s)$ divides four distinct originals, so $M\ge4q$, and a multiple of $q$ exists outside the three used sites. \emph{One bridge.} Write the bridge as $D$, the moduli anchored only at $x$ as $a,b$, and the modulus anchored only at $y$ as $c$. Put $r=\gcd(a,b)$ and $s=\gcd(c,D)$. If both have off-anchor multiples, the argument of the previous paragraph applies: $x$ pays $\lcm(a,b)$, $y$ pays $\lcm(c,D)$, and a collision is repaired by a multiple of $\gcd(r,s)$, which divides four distinct originals. The remaining obstruction is that $s$ has no multiple outside $\{x,y\}$. Since $s\mid D$, this forces $s=D$, so $D\mid c$. The interval contains exactly two multiples of $D$, so $M<3D$, and $c\ne D$ gives $c=2D$. Neither $a$ nor $b$ is divisible by $D$, since the only candidates would be $D$ and $2D$. Put $u=\gcd(a,D)$ and $v=\gcd(b,D)$. Then $u,v\le D/2\le M/4$ (as $2D=c\le M$), $r\le M/3$, and comparing valuations gives \[ \frac{abD}{\lcm(a,b,D)}=r\,\lcm(u,v). \] The anchors pay $\lcm(a,b,D)\cdot c$, so distinct multiples of $r$, $u$ and $v$ off $\{x,y\}$ suffice. Their off-anchor candidate families have at least $1$, $2$ and $2$ elements. A failure of Hall's condition would force the families of $u$ and $v$ to coincide in exactly two points. Since $u$ and $v$ divide $x$ and $y$, their complete sets of interval multiples would coincide, forcing $u=v$. Then $u$ divides the three upper-half originals $a,b,c$, so $u\le M/5$ by Lemma~\ref{lem:upperhalf}, which gives at least three off-anchor multiples of $u$, a contradiction. Hall's theorem gives the representatives. \end{proof} \begin{proof}[Proof of $g(4)\le5$] Form the bipartite graph joining each modulus to its candidate set. Every modulus has a neighbor. A matching of size four gives four sites paying the product, and a matching of size three is completed by Lemma~\ref{lem:onedefect}. If the maximum matching has size at most two, the deficiency version of Hall's theorem gives a set $S$ of moduli with $|S|-|N(S)|\ge2$. Since $N(S)\neq\emptyset$, either three moduli have a common unique anchor, or all four have candidates in a set of at most two sites. In the first case, use Lemma~\ref{lem:oneanchor} if the fourth modulus has no other anchor and Lemma~\ref{lem:triple} otherwise. In the second case, use Lemma~\ref{lem:oneanchor} if the candidate union is a single site, Lemma~\ref{lem:triple} if three moduli share a unique anchor, and Lemma~\ref{lem:twoanchor} otherwise. These cases are exhaustive. \end{proof} \section{A sharp four-modulus instance}\label{sec:sharp4} \begin{proposition}\label{prop:foursharp} Let \[ A_4=\{77,91,139,143\},\qquad I_4=\{5935,5936,\ldots,6077\}. \] No subset of $I_4$ with at most four elements has product divisible by $P(A_4)$, whereas a five-element subset does. \end{proposition} \begin{proof} Here $|I_4|=143=\max A_4$ and $P(A_4)=7^2\cdot11^2\cdot13^2\cdot139$. The only element of $I_4$ divisible by two of $7,11,13$ is $6006=2\cdot3\cdot7\cdot11\cdot13$, since $77\cdot77=5929$, $77\cdot79=6083$, $91\cdot65=5915$, $91\cdot67=6097$, $143\cdot41=5863$ and $143\cdot43=6149$ all lie outside $I_4$. No element of $I_4$ is divisible by $13^2$, because $13^2\cdot35=5915$ and $13^2\cdot36=6084$. A cover must raise the exponents of $7$, $11$ and $13$ to two. If $6006$ is chosen, three further distinct sites are needed for one more factor each of $7$, $11$ and $13$. If $6006$ is not chosen, two sites are needed for $13^2$, and the sites supplying $7$ and $11$ are different from these and from each other. Either way at least four sites divisible by $7$, $11$ or $13$ are used. The only multiple of $139$ in $I_4$ is $5977=43\cdot139$, which is coprime to $7\cdot11\cdot13$. So it is a fifth site, and at least five sites are needed. Conversely \[ B_4=\{5936,5940,5941,5977,6006\}=\{7\cdot848,\ 11\cdot540,\ 13\cdot457,\ 139\cdot43,\ 6006\} \] is a five-element cover. \end{proof} \section{Five moduli: the universal six-site theorem}\label{sec:five} Throughout this section $m_1,\ldots,m_5$ are distinct integers in $[2,M]$. \paragraph{Strategy.} With five moduli a matching argument alone is not enough, so we start from an \emph{anchor assignment}: every modulus is placed on one of its multiples, and the moduli sharing a site must form a divisibility antichain. This restriction is what makes the gcd bounds of Lemma~\ref{lem:antichain} available. We take an assignment with as many occupied sites as possible. With four or five occupied sites the collision repair of Lemma~\ref{lem:onedefect} finishes at once. Otherwise the moduli are grouped on at most three sites, and the group sizes give one of the shapes $(3,1,1)$, $(2,2,1)$, $(3,2)$, $(4,1)$ or $(5)$ (Table~\ref{tab:shapes}). In the shapes $(3,1,1)$, $(2,2,1)$ and $(3,2)$, each anchor pays the lcm of its group. What remains are residual factors, such as the gcd of two moduli in the same group. These are small enough to have spare multiples, and a few additional sites pay them while keeping the total at six. In the two most concentrated shapes, $(4,1)$ and $(5)$, the shared anchor instead pays one prime from each of its moduli, and the remaining proper divisors are placed elsewhere, as in Section~\ref{sec:four}. These shapes are the hardest. With all five moduli on one site, a single exceptional family, five of the six pairwise products of four primes, needs separate treatment (Lemma~\ref{lem:fiveanchor}). With four moduli on one site, a carried gcd may have to pass through several repair sites before it finds a free one (Lemma~\ref{lem:propagation}). \subsection{Antichain anchor assignments} An \emph{anchor assignment} is a map $f$ from the five moduli to sites with $m_i\mid f(m_i)$. It is \emph{admissible} if the moduli sent to any one site form a divisibility antichain. Incomparable moduli may share a site, but comparable moduli may not. \begin{lemma}[A maximal admissible assignment]\label{lem:assign} Admissible assignments exist. Choose one with the largest possible number of occupied sites. If at least four sites are occupied, five sites pay $\prod_im_i$. Otherwise the nonempty fibers have one of the shapes \[ (3,1,1),\quad(2,2,1),\quad(3,2),\quad(4,1),\quad(5). \] \end{lemma} \begin{proof} Assign the moduli in decreasing order. When assigning $d$, forbid an occupied multiple of $d$ only if it carries an earlier modulus that is a multiple of $d$. There are at most $\lfloor M/d\rfloor-1$ such earlier moduli, since they are multiples of $d$ in $(d,M]$, while $I$ contains at least $\lfloor M/d\rfloor$ multiples of $d$. So an allowed multiple exists. Earlier moduli incomparable with $d$ do not matter, so the result is admissible. There are finitely many admissible assignments, so one maximizes the number of occupied sites. If at least four sites are occupied, four distinct moduli sit at distinct multiples, and Lemma~\ref{lem:onedefect} with $t=5$ gives a cover with at most five sites. If at most three sites are occupied, the listed shapes are exactly the partitions of $5$ into at most three parts. \end{proof} Table~\ref{tab:shapes} summarizes how the remaining shapes are handled. \begin{table}[ht] \centering\small \begin{tabular}{@{}cllc@{}} \toprule Occupied sites & Shape & Treated in & Sites used\\ \midrule $\ge4$ & --- & Lemma~\ref{lem:assign} (one-defect repair) & $\le5$\\ $3$ & $(3,1,1)$ & Lemma~\ref{lem:311} & $\le6$\\ $3$ & $(2,2,1)$ & Lemma~\ref{lem:221} & $\le6$\\ $2$ & $(3,2)$ & Lemma~\ref{lem:32} & $\le6$\\ $2$ & $(4,1)$ & Lemmas~\ref{lem:propagation}--\ref{lem:41} & $\le6$\\ $1$ & $(5)$ & Lemma~\ref{lem:fiveanchor} & $\le6$\\ \bottomrule \end{tabular} \caption{Case structure of the proof of $g(5)\le6$.}\label{tab:shapes} \end{table} \begin{lemma}[Antichain gcd bounds]\label{lem:antichain} If $t$ distinct positive multiples of $q$ lie in $[2,M]$ and form a divisibility antichain, then $M\ge(2t-1)q$. In particular, the gcd of an incomparable pair is at most $M/3$, and the gcd of a pairwise incomparable triple is at most $M/5$. If four distinct originals form two incomparable pairs and $q$ divides all four, then $M\ge5q$, even if originals from different pairs are comparable. \end{lemma} \begin{proof} The integers $1,\ldots,N$ split into $\lceil N/2\rceil$ divisibility chains according to their odd parts. An antichain of size $t$ among the quotients, which are at most $\lfloor M/q\rfloor$, therefore requires $\lfloor M/q\rfloor\ge2t-1$. For the last assertion, no quotient equals $1$, since that original would divide its partner. Four distinct quotients that are at least $2$ include one that is at least $5$. \end{proof} \begin{lemma}[Paying a triple residual]\label{lem:tripleresidual} Let $a,b,c$ be pairwise incomparable with a common multiple $x\in I$, and let $F\subseteq I$ contain $x$ and have at most three elements. Suppose each of \[ A=\gcd(a,b),\qquad B=\gcd(a,c),\qquad C=\gcd(b,c) \] has a multiple in $I\setminus F$. Then at most three distinct sites of $I\setminus F$ pay $abc/\lcm(a,b,c)$. \end{lemma} \begin{proof} Put $g=\gcd(a,b,c)$ and write $A=gu$, $B=gv$, $C=gw$. Comparing prime valuations shows that $u,v,w$ are pairwise coprime and \begin{equation}\label{eq:tripleidentity} abc=\lcm(a,b,c)\,g^2uvw. \end{equation} Choose multiples of $A,B,C$ off $F$. If the three sites are distinct, they pay $g^3uvw$. If exactly two sites occur, the shared site pays the lcm of two of $gu,gv,gw$ and the other pays the third label, which gives $g^2uvw$. If all choices coincide at $z$, that site pays $guvw$. Since $M\ge5g$ by Lemma~\ref{lem:antichain}, there is a multiple of $g$ outside $F\cup\{z\}$, which has at most four elements, and it pays the remaining factor. At most three new sites are used in every case. \end{proof} \subsection{Three occupied sites} \begin{lemma}[Shape $(3,1,1)$]\label{lem:311} Suppose $a,b,c$ are pairwise incomparable and assigned to $x$, while $d$ and $e$ are assigned to distinct sites $y,z\ne x$. Then at most six sites suffice. \end{lemma} \begin{proof} Let $A=\gcd(a,b)$, $B=\gcd(a,c)$ and $C=\gcd(b,c)$. If all three have multiples outside $\{x,y,z\}$, apply Lemma~\ref{lem:tripleresidual} with $F=\{x,y,z\}$. The three anchors pay $\lcm(a,b,c)$, $d$ and $e$, and at most three new sites pay the residual $abc/\lcm(a,b,c)$. Otherwise, relabel $a,b,c$ so that $A$ has no multiple outside $\{x,y,z\}$. By Lemma~\ref{lem:antichain}, $M\ge3A$, so $I$ contains at least three multiples of $A$. Hence the multiples of $A$ in $I$ are exactly $x,y,z$, and $M<4A$. Since $a$ and $b$ are incomparable, both are proper multiples of $A$ that are at most $M<4A$, so $\{a,b\}=\{2A,3A\}$. The third original $c$ is not divisible by $A$: a positive multiple of $A$ at most $M$ is $A$, $2A$ or $3A$, which would be either comparable with $a$ or $b$ or equal to one of them. Of $d$ and $e$, at most one is a multiple of $A$ (and then equals $A$). Relabel $d,e$ so that $A\nmid d$; this is harmless because $y$ and $z$ are both multiples of $A$. Put \[ t=\gcd(6A,c),\qquad k=\gcd(A,d). \] We claim that some multiple of $t$ lies outside $\{x,y,z\}$. Note that $t\mid x$, because $\lcm(a,b)=6A$ and $c$ divide $x$, and that $t\le c\le M<4A$. If $t>M/2$, then $t>3A/2$ because $M\ge3A$. The only divisors of $6A$ strictly between $3A/2$ and $4A$ are $2A$ and $3A$. Either would divide $c<4A$ and force $c\in\{2A,3A\}$, a contradiction. Hence $t\le M/2$, and $I$ contains at least two multiples of $t$. If all of them lay among $x,y,z$, two of them would differ by $A$ or $2A$, so $t\mid2A$. The values $t=A$ and $t=2A$ are excluded because $A\nmid c$. If $t\mid A$ and $t\ne A$, then $t\le A/2$. If $t\mid2A$, $t\nmid A$ and $t\ne2A$, then $t\le2A/3$. So $t\le2A/3$, and $M\ge3A$ gives at least four multiples of $t$, again a contradiction. Choose a multiple $w$ of $t$ outside $\{x,y,z\}$. Since $A\nmid d$, we have $k\le A/2$, so $M\ge3A\ge6k$. Choose a multiple $v$ of $k$ outside $\{x,y,z,w\}$. The sites $x,y,z,w,v$ are distinct. Because $A$ divides each of $x,y,z$, these sites pay $\lcm(6A,c)=6Ac/t$, $\lcm(A,d)=Ad/k$ and $e$, respectively; $w$ and $v$ pay $t$ and $k$. The product is divisible by \[ \frac{6Ac}{t}\cdot\frac{Ad}{k}\cdot e\cdot t\cdot k=6A^2cde=abcde. \] So in this blocked case five sites suffice. \end{proof} \begin{lemma}[Shape $(2,2,1)$]\label{lem:221} Suppose incomparable pairs $a,b$ and $c,d$ are assigned to $x$ and $y$, respectively, and $e$ is assigned to a third site $z$. Then at most six sites suffice. \end{lemma} \begin{proof} Put $g=\gcd(a,b)$ and $h=\gcd(c,d)$; both are at most $M/3$ by Lemma~\ref{lem:antichain}. Let $F=\{x,y,z\}$. Suppose first that both have multiples outside $F$, and choose them. If the two sites differ, they pay $gh$. Together with $x,y,z$, which pay $\lcm(a,b)$, $\lcm(c,d)$ and $e$, this covers $abcde$. If both choices coincide at $w$, that site pays $\lcm(g,h)$. Put $q=\gcd(g,h)$. The four originals $a,b,c,d$ are distinct multiples of $q$ forming two incomparable pairs, so $M\ge5q$ by Lemma~\ref{lem:antichain}. A multiple of $q$ outside $F\cup\{w\}$ pays the remaining factor. Suppose instead that one pair gcd, say $g$ after interchanging the pairs, has no multiple outside $F$. Then its multiples in $I$ are exactly the three points of $F$, so $M<4g$ and, as in the previous lemma, $\{a,b\}=\{2g,3g\}$. The other gcd $h$ has a multiple outside $F$. Otherwise its multiples would also be exactly $F$; two three-term arithmetic progressions with the same terms have the same step, so $h=g$, and the incomparable pair $c,d$ would also be $\{2g,3g\}$, contradicting distinctness. Choose a multiple $w\notin F$ of $h$. First suppose that neither $c$ nor $d$ is divisible by $g$. Put $k_c=\gcd(g,c)$, $k_d=\gcd(g,d)$ and $K=\lcm(k_c,k_d)$. Both $k_c$ and $k_d$ are at most $g/2$, so $M\ge3g\ge6k_c,6k_d$. Hence we can choose distinct multiples $u$ of $k_c$ and $v$ of $k_d$ outside $\{x,y,z,w\}$: each label has at least six multiples, and at most five sites are forbidden when the second choice is made. Because $g\mid y$, the anchor $y$ pays $\lcm(c,d,g)$. Comparing valuations gives \[ \lcm(c,d,g)=\frac{cd}{h}\cdot\frac{g}{K}. \] The six sites $x,y,z,w,u,v$ therefore pay \[ 6g\cdot\frac{cdg}{hK}\cdot e\cdot h\cdot k_ck_d=6g^2cde\cdot\frac{k_ck_d}{K}, \] which is divisible by $abcde=6g^2cde$ because $K\mid k_ck_d$. Otherwise one of $c,d$ is a multiple of $g$. The positive multiples of $g$ that are at most $M<4g$ are $g,2g,3g$, and $2g,3g$ are already $a$ and $b$, so we may relabel so that $c=g$. Then neither $d$ nor $e$ is a multiple of $g$. Put $k_d=\gcd(g,d)$ and $k_e=\gcd(g,e)$; both are at most $g/2$, so each has at least six multiples in $I$. Choose distinct multiples of $k_d$ and $k_e$ outside $F$. Since $g$ divides $x$, $y$ and $z$, these anchors pay $6g$, $\lcm(g,d)=gd/k_d$ and $\lcm(g,e)=ge/k_e$. With the two new sites the product is divisible by \[ 6g\cdot\frac{gd}{k_d}\cdot\frac{ge}{k_e}\cdot k_dk_e=6g^3de=abcde. \] In this branch five sites suffice. \end{proof} \subsection{Two occupied sites: the \texorpdfstring{$(3,2)$}{(3,2)} shape} \begin{lemma}[Shape $(3,2)$]\label{lem:32} Suppose $a,b,c$ are pairwise incomparable and assigned to $x$, and $d,e$ are incomparable and assigned to $y\ne x$. Then at most six sites pay $abcde$. \end{lemma} \begin{proof} Let $A=\gcd(a,b)$, $B=\gcd(a,c)$, $C=\gcd(b,c)$ and $h=\gcd(d,e)$. Each of these four labels is at most $M/3$ by Lemma~\ref{lem:antichain}, so it has at least three multiples in $I$ and hence at least one outside $\{x,y\}$. (The labels $A,B,C$ divide $x$ and $h$ divides $y$, but none of them need divide the other anchor.) Call one of $A,B,C$ \emph{tight} if it has exactly one multiple outside $\{x,y\}$. A tight label $R$ has at least three multiples in $I$, so it divides both $x$ and $y$, its multiples in $I$ are exactly $x,y,t$ for some third point $t$, and $M<4R$. Thus $M/4M$, a contradiction. Two equal tight labels would force three distinct originals into the two-element set $\{2R,3R\}$. Choose a multiple $z$ of $h$ outside $\{x,y\}$ which, if a tight label exists, is different from its point $t$. If no such $z$ existed, the multiples of $h$ in $I$, of which there are at least three, would be exactly $x,y,t$, which are also exactly the multiples of $R$. Both sets are three-term arithmetic progressions with common difference equal to the label, so $h=R$. The incomparable pair $d,e$ would then be $\{2R,3R\}$, duplicating two of $a,b,c$. Hence $z$ exists. Each of $A,B,C$ still has a multiple outside $F=\{x,y,z\}$: a label that is not tight has at least two multiples off $\{x,y\}$, and the tight point $t$ was avoided. By Lemma~\ref{lem:tripleresidual}, at most three sites outside $F$ pay $abc/\lcm(a,b,c)$. The anchors $x$ and $y$ pay $\lcm(a,b,c)$ and $\lcm(d,e)$, and $z$ pays $h$. Since $\lcm(d,e)\,h=de$, at most six sites pay $abcde$. \end{proof} \subsection{Prime units and five proper divisors} For an integer $T\ge2$, a \emph{prime unit} of $T$ is a pair $(p,j)$ with $p$ prime and $1\le j\le v_p(T)$. There are $\Omega(T)=\sum_pv_p(T)$ prime units. \begin{lemma}[Prime-unit matching for four originals]\label{lem:fourunits} Let $a_1,\ldots,a_4\ge2$ be pairwise incomparable and $Q=\lcm(a_1,\ldots,a_4)$. Then distinct prime units of $Q$ can be assigned to the four originals so that the unit $(p,j)$ assigned to $a_i$ satisfies $j\le v_p(a_i)$. If $p_i$ is the prime of the unit assigned to $a_i$, then $\prod_{i=1}^4p_i\mid Q$. \end{lemma} \begin{proof} Join $a_i$ to the units $(p,j)$ with $j\le v_p(a_i)$. A subfamily $S$ has exactly $\Omega(\lcm(S))$ neighbors. If $|S|=k\le4$ and $\Omega(\lcm(S))N_2$. Hall's condition therefore holds. \end{proof} \subsection{A single occupied site: the \texorpdfstring{$(5)$}{(5)} shape} \begin{lemma}[Five incomparable originals at one anchor]\label{lem:fiveanchor} If five distinct pairwise incomparable originals have a common multiple $x\in I$, then at most six sites pay their product. \end{lemma} \begin{proof} Let $Q=\lcm(a_1,\ldots,a_5)$, so $Q\mid x$, and form the prime-unit graph of Lemma~\ref{lem:fourunits}. Subfamilies of size at most four satisfy Hall's condition, and the full family does unless $\Omega(Q)\le4$. A divisor lattice with at most three prime units has width at most three, so a failure requires $\Omega(Q)=4$. In that case $Q$ is squarefree. If some prime occurs in $Q$ to exponent $e\ge2$, fixing the exponents of the other primes splits the divisor lattice into at most $2^{4-e}\le4$ chains, which cannot contain five pairwise incomparable divisors. So the divisor lattice is the Boolean lattice on four primes. A five-element antichain in it consists of five of the six two-element subsets. Indeed, a singleton leaves room for at most three further pairwise incomparable subsets of the other three primes; by complementation the same holds for a three-element subset; and the empty and full sets are comparable with everything. After renaming the four primes, the five originals are therefore \begin{equation}\label{eq:exception} \{pq,\,pr,\,ps,\,qr,\,qs\}. \end{equation} If the family is not of this form, Hall's theorem assigns five distinct prime units with primes $p_i$, and $\prod_ip_i\mid Q\mid x$. Put $r_i=a_i/p_i$. Lemma~\ref{lem:fiveproper} gives five distinct multiples of the $r_i$ off $x$. Together with $x$ they pay $\prod_ia_i$ using at most six sites. For the family \eqref{eq:exception}, $\prod_ia_i=p^3q^3r^2s^2$. Apply Lemma~\ref{lem:fiveproper} to the four distinct originals $pq,qr,pr,ps$ with proper divisors $q,q,r,s$. This gives four distinct sites off $x$ that pay $q^2rs$, and $x$ pays $pqrs$. It remains to pay $p^2$ with at most one more site. If $v_p(x)=1$, choose a multiple of $p^2$ in $I$; one exists since $p^2M/2$; \item $2e\notin\{a_1,\ldots,a_4\}$; \item $2e=a_j$ for some $j$, and $r_jM/2$, so $M<4p$. But the larger of the two originals divisible by $p$ is $p\max(q,r)\ge5p$, because $q$ and $r$ are distinct odd primes; this contradicts $M<4p$. Thus $j\ge2$. Give one $2$-unit to $a$, and match $pq,pr,qr$ to the distinct units $p,r,q$. An unused $2$-unit remains in $Q$, and $e=2^{j-1}p$ is even, proving (b). \emph{Case 2: $a=2^j$.} Each of the other three originals has an odd prime factor, because distinct powers of $2$ are comparable. Any two of them have at least two odd prime units in their combined lcm: otherwise both would have the same odd part and differ only by powers of $2$, which is impossible in $(M/2,M]$. If the three together have at least three odd units, they can be matched to odd units. If they have only two, then at least one of them is even, since a two-unit odd integer has no three pairwise incomparable divisors, and one $2$-unit completes the matching. So these three originals use at most one $2$-unit, and we assign another $2$-unit to $a$. If $j\ge3$, a $2$-unit is left over, $e=2^{j-1}$ is even, and (b) holds. If $j=1$, then $a=2$ and four distinct originals cannot lie in $(M/2,M]$. If $j=2$, four distinct originals in $(M/2,M]$ including $4$ force $M=7$ and the originals $\{4,5,6,7\}$. Choosing the primes $2,5,3,7$ gives $P=210$ and $Q=420=2P$, and $e=2$ is even, so (b) holds. \end{proof} \begin{lemma}[Four unique anchors plus one original]\label{lem:41unique} Suppose four pairwise incomparable originals $a_1,\ldots,a_4$ lie in $(M/2,M]$ and have the same unique multiple $x$ in $I$. Let $e$ be a fifth original. Then at most six sites pay all five originals. \end{lemma} \begin{proof} If $e$ has no multiple other than $x$, then $e>M/2$, and the five originals are pairwise incomparable upper-half integers with common multiple $x$; Lemma~\ref{lem:fiveanchor} applies. Otherwise choose a multiple $y\ne x$ of $e$. If $2e$ is one of the $a_i$, take the prime-unit assignment of Lemma~\ref{lem:reserve}; otherwise take any assignment from Lemma~\ref{lem:fourunits}. Let $p_i$ be the assigned primes and $r_i=a_i/p_i$. Since $P=\prod_ip_i$ divides $Q$ and hence $x$, Lemma~\ref{lem:fiveproper} gives four distinct multiples $b_i\ne x$ of the $r_i$. If $e>M/2$, apply Lemma~\ref{lem:propagation}(i). If $e\le M/2$ and $2e$ is not one of the $a_i$, apply Lemma~\ref{lem:propagation}(ii). If $2e=a_j$, then in alternative (a) of Lemma~\ref{lem:reserve} the residual $r_j=2e/p_j\le2e/3M/2$. Apply Lemma~\ref{lem:fourunits} to the four incomparable originals at $x$ (the bridge need not lie in the upper half). The product of the assigned primes divides $x$, and Lemma~\ref{lem:fiveproper} provides four distinct residual sites off $x$. Lemma~\ref{lem:propagation}(i) then gives a cover with at most six sites. \end{proof} \begin{proof}[Proof of $g(5)\le6$] Choose a maximal admissible assignment by Lemma~\ref{lem:assign}. If at least four sites are occupied, that lemma gives at most five sites. With three occupied sites apply Lemma~\ref{lem:311} or Lemma~\ref{lem:221}; with two, Lemma~\ref{lem:32} or Lemma~\ref{lem:41}. With one occupied site, the five originals are pairwise incomparable with a common multiple, and Lemma~\ref{lem:fiveanchor} applies. The shapes are exhaustive, and every construction uses at most six distinct sites. \end{proof} \section{A sharp five-modulus instance}\label{sec:sharp5} \begin{proposition}\label{prop:fivesharp} Let \[ A_5=\{5,10,12,15,20\},\qquad I_5=\{51,52,\ldots,70\}. \] Every product-cover of $A_5$ in $I_5$ has at least six elements, and six suffice. \end{proposition} \begin{proof} Here $|I_5|=20=\max A_5$ and \[ P(A_5)=5\cdot10\cdot12\cdot15\cdot20=180000=2^5\cdot3^2\cdot5^4. \] Exactly four elements of $I_5$ are divisible by $5$, namely $55,60,65,70$, and none is divisible by $25$. So all four belong to every cover. Their product $55\cdot60\cdot65\cdot70=15015000$ contains $2^3\cdot3\cdot5^4$ in the primes $2,3,5$. A fifth element $w$ would have to supply $2^2\cdot3$, that is $12\mid w$. The only multiple of $12$ in $I_5$ is $60$, which is already chosen. Hence no five-element cover exists. On the other hand, $B_5=\{55,60,63,64,65,70\}$ is a cover, since $63=3^2\cdot7$ and $64=2^6$. \end{proof} This instance is due to Chen \cite{Chen708}; the general lower bound $g(n)\ge n+1$ is due to Erd\H{o}s and Sur\'anyi \cite{ErSu1959}. \section{Formal verification}\label{sec:formal} \subsection{The artifact} The formalization is summarized in Table~\ref{tab:artifact}. \begin{table}[ht] \centering\small \begin{tabular}{@{}ll@{}} \toprule Repository & \url{https://github.com/jjtripodo-lgtm/erdos-708-g4-g5}\\ Commit & \href{https://github.com/jjtripodo-lgtm/erdos-708-g4-g5/tree/\commit}{\texttt{\commit}}\\ Toolchain & \texttt{leanprover/lean4:v4.35.0-rc2} \cite{Lean4}\\ Mathlib \cite{Mathlib} & \texttt{065356127b1dc0016f66b7283ce0ce2c4055aa55}\\ Licence & Apache-2.0\\ Registry entry & \palomarid, version~1 \cite{Palomar708}\\ Permitted axioms & \texttt{propext}, \texttt{Quot.sound}, \texttt{Classical.choice}\\ \bottomrule \end{tabular} \caption{The Lean artifact. All dependency revisions are pinned in \texttt{lake-manifest.json}.}\label{tab:artifact} \end{table} The formalization is registered in the Palomar registry as \palomarid, version~1, registered 19 November 2026 at 09:09:46~UTC with trust level ``high'' \cite{Palomar708}. The registry entry records the official verification run \texttt{44442073024} of the Palomar submission workflow (submission \texttt{5h2qqh1ww61l}, verified 02:16:02~UTC). That run checked the four declarations below with the Lean kernel checker and the independent kernels \texttt{nanoda} and \texttt{con-ron}. The automated editorial review recorded no warnings. The registry preserves the source commit and every dependency commit in archived forks under \texttt{PalomarArchive}. \subsection{What is proved} The file \texttt{Challenge.lean} states the theorems with \texttt{sorry} placeholders and imports only Mathlib. \texttt{Solution.lean} proves the same statements, and Palomar's comparator checks that the proved declarations have exactly the Challenge types and depend only on the axioms \texttt{propext}, \texttt{Quot.sound} and \texttt{Classical.choice}. The definitions and statements are reproduced verbatim below, with comments removed: \begin{Verbatim}[fontsize=\footnotesize,frame=single,framesep=2mm] noncomputable def interval (L : ℤ) (M : ℕ) : Finset ℤ := Finset.Icc L (L + (M : ℤ) - 1) def ExactCover (n k : ℕ) : Prop := ∀ m : Fin n → ℕ, Function.Injective m → (∀ i, 2 ≤ m i) → ∀ L : ℤ, ∃ B ⊆ interval L (Finset.univ.sup m), B.card = k ∧ (∏ i, (m i : ℤ)) ∣ ∏ z ∈ B, z def HasCoveringSize (n k : ℕ) : Prop := ∀ A : Finset ℕ, A.card = n → (∀ a ∈ A, 2 ≤ a) → ∀ m : ℕ, 0 < m → ∃ B ⊆ Finset.Ico m (m + A.sup id), B.card ≤ k ∧ (∏ a ∈ A, a) ∣ ∏ b ∈ B, b def IsMinCoveringSize (g : ℕ → ℕ) : Prop := ∀ n, IsLeast {k | HasCoveringSize n k} (g n) theorem erdos_708.variants.four : IsLeast {k | ExactCover 4 k} 5 theorem erdos_708.variants.five : IsLeast {k | ExactCover 5 k} 6 theorem erdos_708.variants.g_four (g : ℕ → ℕ) (hg : IsMinCoveringSize g) : g 4 = 5 theorem erdos_708.variants.g_five (g : ℕ → ℕ) (hg : IsMinCoveringSize g) : g 5 = 6 \end{Verbatim} All four declarations live in the namespace \texttt{Erdos708}. \paragraph{Fidelity.} \texttt{ExactCover n k} is $C^{=}(n,k)$ of Definition~\ref{def:gexact}, with $A$ given as an injective enumeration and intervals starting at any integer. So \texttt{four} and \texttt{five} state $g^{=}(4)=5$ and $g^{=}(5)=6$ unconditionally. \texttt{HasCoveringSize} and \texttt{IsMinCoveringSize} are copied verbatim from the open Formal Conjectures pull request \#4388 \cite{FC4388}. They use the at-most reading with windows of positive integers. Since the upper bounds are proved for every integer start and both extremal instances start at positive integers, this window convention does not affect the values. The statements \texttt{g\_four} and \texttt{g\_five} are implications whose hypothesis requires $g(n)$ to exist for every $n$. That hypothesis is satisfiable: Chen's uniform finite bound $g(n)\le12n$ \cite{Chen708} supplies a covering size for every $n\ge1$. At $n=0$, the empty original set is covered by the empty subset of the empty window, so \texttt{HasCoveringSize 0 0} holds. By well-ordering, every $n$ therefore has a least covering size, defining a function that satisfies \texttt{IsMinCoveringSize}. Chen's bound is not part of this artifact. The same development also proves the unconditional statements \texttt{Erdos708.isLeast\_four} and \texttt{Erdos708.isLeast\_five} (in \texttt{FC4388/Bridge.lean}), namely $\mathrm{IsLeast}\,\{k\mid\texttt{HasCoveringSize}\ 4\ k\}\ 5$ and its analogue for five, but only the four declarations above are compared by the comparator. \subsection{Correspondence with the written proof} Table~\ref{tab:map} lists the Lean counterparts of the main written results. The Lean development follows the same case structure, but several auxiliary results are packaged differently. For example, the Lean lemma \texttt{reserve} returns a general factor split rather than an explicit prime-unit matching, and the Lean $(4,1)$ argument first separates whether $2e$ is an original, rather than whether a bridge exists. Palomar's comparator checks the four final declarations. Their intermediate Lean dependencies are also checked by the kernel, but not every literal written lemma is a separate comparator target or identical in statement to its Lean counterpart. \begin{table}[H] \centering\footnotesize \renewcommand{\arraystretch}{0.92} \begin{tabular}{@{}>{\raggedright}p{0.30\linewidth}>{\raggedright\arraybackslash}p{0.64\linewidth}@{}} \toprule Written result & Lean module and declaration\\ \midrule Theorem~\ref{thm:main}, $n=4$ upper & \path{Erdos708/UniversalUpper.lean}: \texttt{four\_moduli\_five\_cover}\\ Lemma~\ref{lem:onedefect} & \path{Erdos708/OneDefectFinite.lean}: \texttt{one\_defect\_four\_moduli}; \path{Erdos708/Five/MissingOne.lean}: \texttt{missing\_one}\\ Lemma~\ref{lem:proper} & \path{Erdos708/ProperDivisorSDR.lean}: \texttt{proper\_divisor\_sdr\_fin\_four}\\ Lemma~\ref{lem:upperhalf} & \path{Erdos708/UpperHalf.lean}: \texttt{pair\_gcd\_le\_one\_third}, \texttt{triple\_gcd\_le\_one\_fifth}\\ Lemma~\ref{lem:oneanchor} & \path{Erdos708/OneAnchor.lean}: \texttt{one\_anchor\_four\_moduli}\\ Lemma~\ref{lem:triple} & \path{Erdos708/TripleAnchor.lean}: \texttt{triple\_anchor\_four\_moduli}\\ Lemma~\ref{lem:twoanchor} & \path{Erdos708/TwoAnchorCombined.lean}: \texttt{two\_anchor\_combined\_candidate}\\ Proposition~\ref{prop:foursharp} & \path{Erdos708/Sharpness.lean}: \texttt{sharp\_full\_no\_four\_cover}, \texttt{sharp\_example\_exactly\_five}\\ Lemma~\ref{lem:assign} & \path{Erdos708/Five/Assign.lean}: \texttt{exists\_max\_validAssign}\\ Lemma~\ref{lem:antichain} & \path{Erdos708/Five/Antichain.lean}\\ Lemma~\ref{lem:tripleresidual} & \path{Erdos708/Five/Case311.lean}: \texttt{case311\_free} (corresponding payment assembly)\\ Lemma~\ref{lem:311} & \path{Erdos708/Five/Case311.lean}: \texttt{case311}\\ Lemma~\ref{lem:221} & \path{Erdos708/Five/Case221.lean}: \texttt{case221}\\ Lemma~\ref{lem:32} & \path{Erdos708/Five/Case32.lean}: \texttt{case32}\\ Lemma~\ref{lem:fourunits} & \path{Erdos708/SmallAntichainUnion.lean}: \texttt{small\_pairwise\_incomparable\_biUnion\_card\_candidate}; matching in \path{Erdos708/Five/Case41.lean}\\ Lemma~\ref{lem:fiveproper} & \path{Erdos708/Five/SDR5.lean}: \texttt{proper\_divisor\_sdr\_fin\_five}\\ Lemma~\ref{lem:fiveanchor} & \path{Erdos708/Five/Case5.lean}: \texttt{case5\_holds}\\ Lemma~\ref{lem:propagation} & \path{Erdos708/Five/Case41Core.lean}: \texttt{case41\_core}\\ Lemma~\ref{lem:reserve} & \path{Erdos708/Five/Case41Reserve.lean}: \texttt{reserve}\\ Lemmas~\ref{lem:41unique}, \ref{lem:41} & \path{Erdos708/Five/Case41.lean}: \texttt{case41\_holds}\\ Theorem~\ref{thm:main}, $n=5$ upper & \path{Erdos708/Five/Final.lean}: \texttt{five\_moduli\_six\_cover}\\ Proposition~\ref{prop:fivesharp} & \path{Erdos708/Five/Sharpness.lean}: \texttt{sharp5\_no\_five\_cover}, \texttt{sharp5\_six\_cover}\\ Lemma~\ref{lem:padding}, Corollary~\ref{cor:exact} & \path{Adapter.lean}: \texttt{exact\_of\_bound}; \path{Solution.lean}: \texttt{least\_exact}\\ At-most statements & \path{FC4388/Bridge.lean}: \texttt{isLeast\_four}, \texttt{isLeast\_five}\\ \bottomrule \end{tabular} \caption{Written results and structural Lean counterparts at commit \texttt{00fbb23c}; rows are not claims of literal theorem-type identity. Declarations in \texttt{Erdos708/Five/} are in the namespace \texttt{Erdos708.Five}, those in \texttt{Adapter.lean} in \texttt{Erdos708Finite}, and the others in \texttt{Erdos708}.}\label{tab:map} \end{table} \subsection{Reproduction} With \texttt{elan} installed, the following commands from the repository README rebuild the proofs and rerun the comparator (the comparator requires \texttt{bubblewrap}): \begingroup\small \begin{verbatim} lake exe cache get lake build Challenge Solution lake comparator --config comparator.json lake comparator --config comparator.json --paranoid \end{verbatim} \endgroup Before release, the paranoid comparator was also run locally on the same toolchain, where the proofs were accepted by Lean's default and paranoid checks and by the external kernels \texttt{lean4lean}, \texttt{nanoda}, \texttt{con-leche} and \texttt{con-ron}. Mutation tests confirmed that altered statements are rejected: changing the value $5$ to $4$ in \texttt{four}, weakening $2\le m_i$ to $1\le m_i$ in \texttt{ExactCover}, changing \texttt{g 4 = 5} to \texttt{g 4 = 6}, and changing $0