---

Subject and independence

Role. Independent reviewer working in a fresh context from the assignment alone; took no part in writing the reviewed section, the lead, the Tools section or the reconstruction pages. Charge: refutation.

Subject. The page wiki/research/erdos_1219/connections.md as it stood at 2026-09-28T13:50:48Z, section ## Hajnal's Conjecture 1A: triples, lines 286--445 (the H2 at line 286 to the line before the H2 ## Problem 1218 at line 446), and the lead page wiki/research/leads/triple_canonization_for_hajnal_conjecture/_index.md (151 lines, whole page), both as of that time.

Reading. The section and the lead in full, every sentence. The page's Scope paragraph (lines 19--37) and its Tools section (lines 39--108) as inputs: the interface (T1) was confirmed at each use (hypotheses on $\kappa$, $\mu(i)$, $\lambda_i$, $A_i$, the arity of $F$, the conclusions (1A), (1B), (2)); (T2) and (T3) are not used by the section and were not checked. The three reconstruction pages wiki/research/erdos_1219/lemma_1_1_reconstruction.md, wiki/research/erdos_1219/theorem_1_2_reconstruction.md and wiki/research/erdos_1219/corollary_1_3_reconstruction.md in full, as premises at their standing; the section's uses of clauses (1A), (1B), (2), of the argument for clause (3), of the proof-internal first thinning, and of Steps 1, 2, 5 and 7 of the theorem page were checked against their stated hypotheses. The statement paragraph of wiki/problems/set_theory/E1219/_index.md (lines 17--23; the page has no Statement heading; its Current assessment, Progress and Known Results sections were not read). The result page library/set_theory/shelah_1975_notes_partition_calculus/conjecture_1a.md and the provenance and read-status paragraphs of the Shelah library card. The held Shelah scan, PDF pp. 2--5 (printed pp. 1258--1261), on page images rendered at 130 dpi: the statement of Lemma 1.1 with its Remark, the proof on pp. 1259--1260, Theorem 1.2 with its proof, Corollary 1.3 with its Remark, and Conjecture 1A with its Remark. The held Komjáth survey, PDF pp. 2--3 (printed pp. 419--420), text layer: Problems 3, 4 and 5 with their commentary. The wiki pages named in the assignment.

Exposures. None to excluded content. A directory listing made to locate the target path showed the file names of other records under wiki/research/erdos_1219/evidence/verify/; none was opened. A working tree status listing showed the names of untracked review files in a sibling research folder; none was opened.

Restatement

Throughout, H is the hypothesis $\aleph_\omega<2^{\aleph_{n(0)}}<2^{\aleph_{n(1)}}<\cdots$ for a sequence $(n(k))_{k<\omega}$ of natural numbers, $\chi=\sum_{n<\omega}2^{\aleph_n}$, and a coloring is a map $f:[\chi]^3\to\{0,1\}$. The blocks are those of the theorem page at $\lambda=\aleph_\omega$, $\kappa=\omega$: cardinals $\mu(i)<\aleph_\omega$ ($i<\omega$), strictly increasing and cofinal in $\aleph_\omega$, each above $\aleph_{n(0)}$, with $2^{\mu(i)}$ strictly increasing and $2^{\mu(i)}>\aleph_\omega$; $\lambda_i=(2^{\mu(i)})^+$; and a partition $\chi=\bigsqcup_{i<\omega}A_i$ with $|A_i|=\lambda_i$. The pattern of a triple meeting the blocks $i\le j\le k$ is $(i,j,k)$. For a family of sets $B_i\subseteq A_i$, $E_j=\bigcup_{i<j}B_i$.

• S1 (target; not a claim). Conjecture 1A: under H, every coloring $f$ has $X\subseteq\chi$ with $|X|=\aleph_\omega$ and $f\equiv0$ on $[X]^3$, or $Q\subseteq\chi$ with $|Q|=4$ and $f\equiv1$ on $[Q]^3$. The section attributes to the Remark on printed p. 1261 the unproved assertion $\chi\not\to(\aleph_\omega,5)^3$ and to the survey, p. 419, the record that the conjecture is unproven.
• S2 (the canonization statement; a definition). A coloring $f$ satisfies (C) when there are $S_i,T_i\subseteq A_i$ ($i<\omega$) with $|S_i|=\mu(i)$, $f\equiv0$ on $[S_i]^3$, $T_i$ a three-element set with $f(T_i)=1$, such that with $B_i=S_i\cup T_i$: (P1) for all $i<j<k$ the value $f(\{a,b,c\})$ is the same for all $a\in S_i$, $b\in S_j$, $c\in S_k$; (P2) for all $i<j$ the value $f(\{a,a',b\})$ is the same for all $\{a,a'\}\in[B_i]^2$ and $b\in B_j$; (P3) for all $i<j$ the value $f(\{a,b,b'\})$ is the same for all $a\in B_i$ and $\{b,b'\}\in[B_j]^2$.
• S3 (endgame lemma). Under H, with the blocks fixed: if every coloring $f$ satisfying (i) no four-element $Q$ has $f\equiv1$ on $[Q]^3$, and (ii) no $i<\omega$ and no $X\subseteq A_i$ with $|X|=\aleph_\omega$ has $f\equiv0$ on $[X]^3$, satisfies (C), then Conjecture 1A holds. Imported: $\omega\to(\omega)^3_2$ and $\omega\to(\omega)^2_2$.
• S4 (remark on the triples $T_i$). Let $f_0$ give color $1$ exactly to the triples of pattern $(i,i,j)$ with $i<j$. Then (a) $f_0$ has no four-element set homogeneous in color $1$; (b) every set of size $\aleph_\omega$ homogeneous in color $0$ lies, up to finitely many points, inside one block; (c) $f_0$ "shows that the case $\eta_2=1$, $\varepsilon=0$ cannot be handled without the $T_i$".
• S5 (application of the lemma). Let $\tilde f:A^3\to2$ be the symmetric extension of $f$, $0$ on tuples with a repeated entry, and let $P_i(\langle B_l:l\le i\rangle,\langle a_j:i<j<\omega\rangle)$ hold when $B_i=S\cup T$ with $|S|=\mu(i)$, $f\equiv0$ on $[S]^3$ and $T$ a triple of color $1$ (the section adds "together with whatever (P2) turns out to need"). For a coloring with (i) and (ii), and given the imported relation $(2^\mu)^+\to(\mu^+,4)^3$ for infinite $\mu$, the realizability hypothesis (H) holds for these $P_i$, so (T1) applies and Lemma 1.1 yields $a^*_i\in A_i$ and $B_i\subseteq A_i$ with (1A), (1B) and (2).
• S6 (P1). For $i<j<k$, $a\in B_i$, $b,b'\in B_j$, $c,c'\in B_k$: $f(\{a,b,c\})=f(\{a,b',c'\})$. The map sending $a\in S_i$ to the sequence $\langle f(\{a,b,c\}):i<j<k\rangle$ has at most $2^{\aleph_0}$ values; when $2^{\aleph_0}<\mu(i)$ for every $i$, replacing each $S_i$ by a fiber of size $\mu(i)$ and keeping $T_i$ gives (P1) while (1A), (1B) and (2) survive; $2^{\aleph_0}<\mu(i)$ for every $i$ can be arranged through the choice of $\mu_0$ in Step 1 exactly when $2^{\aleph_0}<\aleph_\omega$, and when $2^{\aleph_0}>\aleph_\omega$ the fiber argument fails.
• S7 (P2, first half). For $i<j$, $\{a,a'\}\in[B_i]^2$ and $b\in B_j$: $f(\{a,a',b\})=f(\{a,a',a^*_j\})$. Hence (P2) holds if and only if, for every $i$, the coloring $c_i(\{a,a'\})=\langle f(\{a,a',a^*_j\}):j>i\rangle$ of $[B_i]^2$, with at most $2^{\aleph_0}$ colors, is constant; the lemma's thinnings do not provide this, so it must be a conjunct of $P_i$, realized inside every $C\subseteq A_i$ with $|C|=\lambda_i$.
• S8 (P2, the Erdős--Rado discussion). When $2^{\aleph_0}\le\mu(i)$, the imported relation $(2^\mu)^+\to(\mu^+)^2_\mu$ gives $C'\subseteq C$ with $|C'|=\mu(i)^+$ on which $c_i$ is constant; (S8a) inside $C'$ the relation $(2^\mu)^+\to(\mu^+,4)^3$ "cannot be applied again ($|C'|$ is not of that form)"; (S8b) a triple of color $1$ inside $C'$ is not guaranteed; a $c_i$-homogeneous set of size $\lambda_i$ cannot be asked because $\lambda_i$ is a successor cardinal; so the route needs the block statement (BS2): in every $C\subseteq A_i$ of size $\lambda_i$, for every coloring of $[C]^2$ with at most $2^{\aleph_0}$ colors, there is $S\cup T$ as in (C) on which that coloring is constant, or $f$ has a four-element set of color $1$; (BS2) is left open.
• S9 (P3, what the lemma gives). (S9a) No clause of Lemma 1.1 relates $f(\{a,b,b'\})$, with $\{b,b'\}\subseteq B_j$ and $a\in E_j$, to any other value. (S9b) "(P3) says that $d_j(\{b,b'\})=\langle f(\{a,b,b'\}):a\in E_j\rangle$, a coloring of $[B_j]^2$ with at most $2^{|E_j|}=2^{\mu(j-1)}$ colors, is constant."
• S10 (P3, the counting obstruction). For $j\ge1$, $2^{\mu(j-1)}\ge2^{\aleph_{n(0)}}>\aleph_\omega>\mu(j)$; the Erdős--Rado relation at the block size $\lambda_j$ would need $2^{\mu(j)}\ge2^{2^{\mu(j-1)}}$; when $2^{\mu(j)}<2^{2^{\mu(j-1)}}$ there is a coloring of $[\lambda_j]^2$ with $2^{\mu(j-1)}$ colors and no homogeneous triple; choosing smaller earlier blocks does not remove the obstruction, because the $\mu(i)$ are cofinal in $\aleph_\omega$; whether the particular $d_j$ induced by an $f$ with (i) admits a homogeneous $S_j\cup T_j$ is the open gap.
• S11 (falsification checks). The endgame uses only the values on the $B_i$ and the two Ramsey theorems; its cases are exhaustive; the block sizes are those of Theorem 1.2, so (T1) applies; the Remark after Lemma 1.1 was not read; the three-color form of Theorem 1.2 gives nothing for the triple relation.
• S12 (conditional deductions of the next investigation). (S12a) If every $f$ with (i) and (ii) forces, in every full-size subset of $A_j$, a set $S_j\cup T_j$ on which $d_j$ is constant, then, with (BS2), the conjecture holds under $2^{\aleph_0}<\aleph_\omega$ by the endgame lemma. (S12b) A coloring $f$ defeating (P3) in every choice of $S_j\cup T_j$ "shows that a proof of the conjecture cannot pass through canonization over the lower blocks"; the lead: "shows that no proof passes through canonization over the lower blocks".
• S13 (outcome sentence). The endgame lemma reduces the conjecture to (C); the lemma supplies (P1) under $2^{\aleph_0}<\aleph_\omega$ and half of (P2), and nothing of (P3).

The lead page restates S1 (Target and status evidence), S2 and S3 (Route and established inputs), S6, S7 and S9a (What the lemma supplies), the imported inputs (Provisional premises), S9b and S10 (The consequential gap), and S12 with the removal of $2^{\aleph_0}<\aleph_\omega$ as a later step (Next step). Its frontmatter marks the lead a candidate, unreviewed, and its closing sentence disclaims any status change.

Checklist

• Quantifiers and scope. Defect found. S9b drops the quantifier over $a\in B_i$: (P3) demands constancy in the pair $\{b,b'\}$ and constancy in $a$ within each lower block, while the constancy of $d_j$ gives only the first (F1). (C) asks (P3) for $a\in B_i\supseteq T_i$, although only $a\in S_i$ is used by the endgame and only that much is deliverable by thinning (F1). S12b's "no proof passes" is a universal statement about proofs that the negative outcome does not support (F2). The (P3) question also depends on the admissible lower sequence through $E_j$, which the wording "induced by an $f$" hides (F7). The endgame lemma's quantifiers are correct.
• Circularity. None. (C) is a structural sufficient condition; the endgame lemma proves the implication from (C) to the target without using the target, and nothing in the section derives (C) from the conjecture.
• Model and convention changes. Pass. The passage from $f$ on $[\chi]^3$ to the symmetric $\tilde f$ on $A^3$ and the identification of $\chi$ with the union of the blocks are proved transfers (Step 2(c), (d) of the theorem page); the pattern convention is used consistently; the reading of $2^{\chi+\kappa}<\operatorname{cf}\mu(i)$ as $2^{\aleph_0}<\mu(i)$ is correct because every $\mu(i)$ is an $\aleph_m$ with $m<\omega$, hence regular.
• Finite and statistical overreach. Pass; nothing finite or statistical is used.
• Uniformity. Pass. The bound $2^{\aleph_0}$ on the fiber maps in S6 and S7 does not depend on $i$; the requirement $2^{\aleph_0}<\mu(i)$ is stated for every $i$ and met by one choice of $\mu_0$; the color count $2^{\mu(j-1)}$ in S10 depends on $j$ as stated, and the obstruction is argued for every $j\ge1$.
• Extremal conclusions. Inapplicable; no infimum, supremum or sharpness claim is made.
• Consequences and composition. Defects found. S12a claims a conclusion whose bridge from the constancy of $d_j$ to (P3) is missing (F1). S8a gives a false reason for a true consequence (F3). S13's "the lemma supplies (P1)" also consumes the imported triple relation through (H), which the Outcome sentence does not carry (F8). Every other "hence" was checked separately and follows.
• Computation. Inapplicable; no computation is used or claimed.
• Reproduction. Inapplicable; no rerun instruction or coverage claim is made.
• Source and verdict fidelity. Pass with two unverifiable items. The Remark on printed p. 1261, the survey's record on p. 419, the note $2^\kappa\not\to(3)^2_\kappa$ under Problem 5 on p. 420, and the Remark after Lemma 1.1 on p. 1258 read on the held pages as the section reports them. The attribution of $(2^\mu)^+\to(\mu^+,4)^3$ to Erdős, Hajnal and Rado (1965) cannot be checked in the read set, and the pages say so ("not held"). The sentences citing a search "recorded on the problem page (2026-09-27)" and an "assessment dated 2026-09-27" refer to excluded text and were not checked. The standing sentences (author-recorded, unreviewed, no status change, no tier) claim nothing beyond author-recorded.

Weakest steps

W1: the reduction of (P3) to the constancy of $d_j$ (S9b), and the bridge in S12a. Fix $i<j$. (P3) at $(i,j)$ says: there is $h_1(i,j)\in\{0,1\}$ with $f(\{a,b,b'\})=h_1(i,j)$ for all $a\in B_i$ and all $\{b,b'\}\in[B_j]^2$. Write $v_{b,b'}:E_j\to2$ for $a\mapsto f(\{a,b,b'\})$, so $d_j(\{b,b'\})=v_{b,b'}$. Then (P3) for all $i<j$ is the conjunction of (P3a) $d_j$ is constant on $[B_j]^2$, with common value $v_j$, and (P3b) $v_j$ is constant on each $B_i$, $i<j$. Constancy of $d_j$ is (P3a) alone. (P3a) does not imply (P3b): if $\varphi:\chi\to2$ takes both values on some $B_i$ and $f(\{a,b,b'\})$ is $\varphi(a)$ on every triple of pattern $(i,j,j)$, $i<j$, whatever $f$ does elsewhere, then $d_j(\{b,b'\})=\varphi\restriction E_j$ for every pair, so $d_j$ is constant for every choice of $B_j$, while (P3) fails at $(i,j)$. Now compose with the lemma. Neither (P3a) nor (P3b) is provided by any clause of Lemma 1.1 (S9a is correct, and (P3b) is equally uncovered: the lemma canonizes an element of $B_i$ against one point $a^*_j$ of each higher block, never against a pair). If (P3a) is obtained, as S12a supposes, (P3b) for $a\in S_i$ follows by an extra fiber step: the map $a\mapsto\langle v_j(a):j>i\rangle$ on $S_i$ has at most $2^{\aleph_0}$ values, so under $2^{\aleph_0}<\mu(i)$ a fiber of size $\mu(i)$ exists, and it may be taken jointly with the (P1) fiber of S6. (P3b) for $a\in T_i$ cannot be obtained by thinning: $T_i$ is three fixed points chosen at stage $i$, before any $B_j$ with $j>i$ exists, from information that contains one point of each higher block and therefore cannot see any pair in $B_j$. The endgame lemma uses (P3) only for $a\in S_i$: in the case $\eta_1=1$ the single lower point may be taken in $S_i$, and the final case uses the $S_i$ alone. So (C) with (P3) restricted to $a\in S_i$ still supports the endgame, and S12a holds for that restricted (C) after the extra fiber step, under the hypothesis $2^{\aleph_0}<\aleph_\omega$ it already carries. As written, with (P3) over $B_i$ and with the bridge absent, S12a does not follow.

W2: the (P1) transfer through the argument of clause (3) (S6). By (1B) at the blocks $(j,k)$ with the one-element sequence $(a)$ from $B_i$, $i<j$: $\tilde f(b,c,a)=\tilde f(b',c',a)$ for $b,b'\in B_j$, $c,c'\in B_k$, which is the joint independence of $f(\{a,b,c\})$ from $b$ and $c$. The residual map $H_i$ on $S_i$ is indexed by the countably many pairs $(j,k)$ with $i<j<k<\omega$ and takes values in $2$, so it has at most $2^{\aleph_0}$ values. Since $\mu(i)=\aleph_m$ for some $m<\omega$, $\mu(i)$ is regular, and if $2^{\aleph_0}<\mu(i)$ then $S_i$, a union of at most $2^{\aleph_0}<\operatorname{cf}\mu(i)$ fibers, has a fiber of size $\mu(i)$; call it $S'_i$ and put $B'_i=S'_i\cup T_i$. (1A) and (1B) are universal over the $B_i$ and survive; $P_i$ holds for $B'_i$ because it depends on $B_i$ alone and $S'_i\cup T_i$ has the required shape; the value $H_i(a)$ was independent of the choice of $b\in B_j$, $c\in B_k$, so it is unchanged on $B'_j$, $B'_k$; hence (P1) holds on the $S'_i$. The condition $2^{\aleph_0}<\mu(i)$ for all $i$ is arranged by taking $\mu_0$ above $2^{\aleph_0}$ in Step 1 when $2^{\aleph_0}<\aleph_\omega$; when $2^{\aleph_0}>\aleph_\omega$ no $\mu(i)<\aleph_\omega$ exceeds $2^{\aleph_0}$; $2^{\aleph_0}=\aleph_\omega$ is excluded by the cofinality of $2^{\aleph_0}$ being uncountable, so "exactly when" is right. The step does not invoke clause (3) as stated, whose heredity hypothesis the property fails (dropping a point of $T$ from a same-size subset destroys the color-$1$ triple); it reuses the reconstruction's fiber argument with the thinning confined to $S_i$, which is sound (F4).

W3: the endgame lemma (S3). Let $f$ satisfy (i) and (ii) and (C). Ramsey for triples on $g:[\omega]^3\to2$ gives an infinite $I$ with constant value $\varepsilon$; Ramsey for pairs, twice, gives an infinite $I'\subseteq I$ with $h_1\equiv\eta_1$ and $h_2\equiv\eta_2$ on $[I']^2$. If $\varepsilon=1$: four indices of $I'$ and one point of each $S$ give a four-element set all of whose triples have three distinct block indices in $I'\subseteq I$, hence color $1$. If $\eta_2=1$: for $i<j$ in $I'$, $T_i\cup\{b\}$ with $b\in B_j$ has the triple $T_i$ of color $1$ and three triples with two points in $T_i\subseteq B_i$ and one in $B_j$, of color $h_2(i,j)=1$ by (P2) as quantified over $[B_i]^2$. If $\eta_1=1$: one point $a\in B_i$ and $T_j$ give $T_j$ of color $1$ and three triples with one point in $B_i$ and a pair from $T_j\subseteq B_j$, of color $h_1(i,j)=1$ by (P3). Otherwise $\varepsilon=\eta_1=\eta_2=0$ and $X=\bigcup_{i\in I'}S_i$ is homogeneous in color $0$: a triple inside one $S_i$ has color $0$ by the choice of $S_i$, and a triple across blocks has pattern $(i,i,j)$, $(i,j,j)$ or $(i,j,k)$ with $i<j<k$ in $I'$ and color $\eta_2$, $\eta_1$ or $\varepsilon$. The $S_i$ are disjoint and $I'$ is unbounded in $\omega$, so $|X|=\sum_{i\in I'}\mu(i)=\aleph_\omega$. The case split is exhaustive, and colorings without (i) or (ii) satisfy the conjecture directly. The lemma is correct. Its proof needs (P2) only for $b\in S_j$ and (P3) only for $a\in S_i$, which is what W1 uses.

Strongest attack

• S3. Attempted: making the four-element sets degenerate by letting $T_i$ meet $S_i$ (irrelevant, as the blocks are disjoint and the sets used have four points in two or four blocks); using (P2) or (P3) at pairs from $T$ (available, since (C) quantifies over $[B_i]^2$ and $[B_j]^2$); making $|X|<\aleph_\omega$ (impossible, $I'$ is infinite and the $\mu(i)$ are cofinal); finding the conjecture inside the hypothesis (it is not: (C) is a statement about sets in blocks, and a counterexample to the conjecture with (i) and (ii) would simply fail (C)). All failed.
• S4. (a) and (b) were re-derived: a four-element set homogeneous in color $1$ would need every triple to have exactly two points in its lowest block, which no distribution of four points over blocks achieves; a set homogeneous in color $0$ with two points in a block $i$ has no point above block $i$ and at most one point in each block below it. For (c), the attack is that $f_0$ violates hypothesis (ii) (each block is homogeneous in color $0$), so it lies outside the endgame lemma and defeats no implication the section states; what it shows is that the canonical values on the $S_i$ alone are compatible with both outcomes in the case $\eta_2=1$, $\varepsilon=0$. As a statement about the proof mechanism, (c) survives (F5).
• S5. Attempted: breaking (H). Inside every $C\subseteq A_\alpha$ with $|C|=\lambda_\alpha=(2^{\mu(\alpha)})^+$, the imported relation with $\mu=\mu(\alpha)$ and (i) give a set of size $\mu(\alpha)^+$ homogeneous in color $0$, which contains one of size $\mu(\alpha)$; (ii) gives a triple of color $1$ because $|C|>\aleph_\omega$; the union has size $\mu(\alpha)$. The lemma's remaining hypotheses are those (T1) verifies. The types over a set include patterns with repeated entries, on which $\tilde f$ is $0$; harmless. Failed, conditionally on the imported relation.
• S6. See W2. Attempted: an uncountable index set for $H_i$ (no, the pairs $(j,k)$ are countable); a singular $\mu(i)$ (no, below $\aleph_\omega$ every infinite cardinal is regular); loss of (2) under thinning (no, $P_i$ depends on $B_i$ alone). Failed.
• S7. Attempted: the link between $b\in B_j$ and $a^*_j$ is not among the lemma's conclusions (1A), (1B), (2); (1B) relates one point of $B_i$ to $a^*_j$, not a pair. The section names the proof-internal first thinning of the lemma page, which gives $b$ and $a^*_j$ the same type over $E_j$ and hence $\tilde f(b,a,a')=\tilde f(a^*_j,a,a')$; the reconstruction states this as its labeled fact (T1). The "if and only if" then holds, and the count of colors is $2^{\aleph_0}$. Failed (F6).
• S8a. Refuted-as-stated. Under $2^{\aleph_0}=\aleph_2$, $n(0)=1$ and $2^{\aleph_2}>2^{\aleph_1}$, Step 1 gives $\mu(0)=\aleph_2$, the hypothesis $2^{\aleph_0}\le\mu(0)$ of the sentence holds, and $|C'|=\aleph_3=(2^{\aleph_0})^+$ is of the form $(2^\nu)^+$ with $\nu=\aleph_0$, so the relation applies again. The consequence survives for a different reason: whenever $\mu(i)=2^\nu$, then $\nu<\aleph_{n(0)}<\mu(i)$ (otherwise $2^\nu>\aleph_\omega$), so the relation yields a set of color $0$ of size $\nu^+<\mu(i)$, and S8b already defeats the approach.
• S8b. Attempted: a triple of color $1$ inside $C'$ is guaranteed by (ii) only for sets of size at least $\aleph_\omega$, and $\mu(i)^+<\aleph_\omega$; a homogeneous set of size $\lambda_i$ for a two-valued coloring of $[\lambda_i]^2$ is refuted by the Sierpiński coloring restricted from $2^{2^{\mu(i)}}\ge\lambda_i$, as $\lambda_i$ is a successor. Failed.
• S9a. Attempted: each of (1A), (1B), (3) was checked for a tuple with two entries from one block and one from a lower block; none covers it. The proof-internal second thinning relates $f(\{a,b,a^*_j\})$ to $f(\{a,b',a^*_j\})$ for $b,b'\in B_j$, a single pair type, not $[B_j]^2$. Failed.
• S9b. Refuted-as-stated; witness in W1.
• S10. Re-derived: $|E_j|=\sum_{i<j}\mu(i)=\mu(j-1)$ for $j\ge1$ since each $|B_i|=\mu(i)$; the Erdős--Rado relation $(2^\theta)^+\to(\theta^+)^2_\theta$ with $\theta=2^{\mu(j-1)}$ needs $\lambda_j\ge(2^\theta)^+$, that is $2^{\mu(j)}\ge2^{2^{\mu(j-1)}}$; otherwise $\lambda_j\le2^\theta$, and the coloring of $[{}^\theta2]^2$ by the least coordinate where two functions differ has no homogeneous triple (three functions cannot be pairwise distinct at one coordinate with two values), which is $2^\theta\not\to(3)^2_\theta$ restricted to $\lambda_j$ points; a block $i$ with $\mu(i)\ge\aleph_{n(0)}$ has $2^{\mu(i)}>\aleph_\omega$, and all but finitely many blocks are such. Attempted: the bound $2^{\mu(j-1)}$ is an upper bound on the colors of $d_j$, which the section already concedes by leaving the induced $d_j$ open. Failed.
• S11. Re-derived with W3 and (T1). Failed.
• S12a. Proof defective; statement unresolved; see W1 and F1.
• S12b. Proof defective; statement unresolved; a negative outcome refutes (H) for the property that carries (P3), hence (C) as the route's sufficient condition for that $f$ and those blocks. It does not exclude a proof through a different canonical form (constancy on the $S_i$ after thinning, constancy up to a small exceptional set, a different block decomposition or a different parameter set), so the universal "no proof passes through canonization over the lower blocks" is unsupported (F2).
• S13. Attempted: "supplies (P1)" without its imported input; the Outcome sentence carries only $2^{\aleph_0}<\aleph_\omega$, while the realizability (H) behind (P1) also consumes $(2^\mu)^+\to(\mu^+,4)^3$, which the section's next-investigation paragraph and the lead's provisional premises do carry. Survives as author-recorded with that qualification (F8).

Premises

• (T1), Tools section of the connections page, lines 46--61. Interface: $\kappa$ infinite regular; $\mu(i)$ infinite, strictly increasing, $\mu(0)\ge\kappa$, $2^{\mu(i)}$ strictly increasing; $\lambda_i=(2^{\mu(i)})^+$; $A_i$ pairwise disjoint of size $\lambda_i$; $F:A^n\to2$ of finitely many places; then Lemma 1.1 applies with $\chi=2$, $F_0=F_1=F$ and any $P_\alpha$ satisfying (H), yielding (1A), (1B), (2) with $|B_i|\le\mu(i)$. Its proof (Steps 2(a) and 5 of the theorem page) is another reviewer's subject; here confirmed: $\kappa=\omega$, $\mu(0)>\aleph_{n(0)}\ge\aleph_0$, $n=3$, the blocks of Step 2.
• Lemma 1.1 reconstruction (premise at its standing). Used: clauses (1A), (1B), (2) as stated on its Statement section; the labeled proof-internal fact (T1) (first thinning: every $b\in B_\alpha$ has the type of $a^*_\alpha$ over $E_\alpha$); the fiber argument of its Clause (3) section with its reading of the hypotheses. Read in full.
• Theorem 1.2 reconstruction (premise at its standing). Used: Step 1 (the $\mu(i)$, with $\mu(i)>\rho_i\ge\mu_0$ and $\mu(i)\ge\kappa$), Step 2 (the blocks and $\chi=\bigsqcup A_i$), Step 5 (the hypotheses of the lemma), Step 7 (the size of a union of the $S_i$ over an unbounded index set). Read in full.
• Corollary 1.3 reconstruction (premise at its standing). Used: the hypotheses of Theorem 1.2 at $\lambda=\aleph_\omega$ under H, with $\mu_0=\aleph_{n(0)}$, and $\chi=\sum_{n<\omega}2^{\aleph_n}$. Read in full.
• Conjecture 1A result page (conjecture_1a). Statement and Remark as recorded; matches the held page image, printed p. 1261.
• Shelah (1975), held scan (library card), PDF pp. 2--5 on page images. Lemma 1.1 with its Remark (p. 1258), the proof (pp. 1259--1260) including the one-sentence treatment of (3), Theorem 1.2, Corollary 1.3 and its Remark (p. 1260), Conjecture 1A and its Remark (p. 1261). Read at statement depth for the statements the section cites; the proofs were read to confirm the reconstruction's labels, not re-verified.
• Komjáth (2025), held PDF (library card), printed pp. 419--420 in the text layer: Problem 3 with the record of Hajnal's conjecture as unproven; Problem 5 with "$2^\kappa\not\to(3)^2_\kappa$ holds in general".
• Imported, not held. Ramsey's theorem $\omega\to(\omega)^3_2$ and $\omega\to(\omega)^2_2$ (standard; used in S3). Erdős--Rado $(2^\mu)^+\to(\mu^+)^2_\mu$ for infinite $\mu$ (standard; used in S8 and S10 as a known relation). The unbalanced triple relation $(2^\mu)^+\to(\mu^+,4)^3$ for infinite $\mu$ (used in S5; its attribution to Erdős, Hajnal and Rado 1965 is not checkable here and is flagged on both pages as to be located; the Remark on p. 1261, which asserts that every previously known case of $\lambda\to(\mu,\mu)^2$ also gives $\lambda\to(\mu,4)^3$, is consistent with it but proves nothing). $2^\theta\not\to(3)^2_\theta$ (held in the survey and re-derived above).
• Explicit assumptions of the route. H; the blocks of Step 2; $2^{\aleph_0}<\aleph_\omega$ for S6 and the Erdős--Rado step of S8; the block statement (BS2); the (P3) question. No batch acceptance order applies.

Findings

• F1 (required). Location: connections page, (P3) bullet, "(P3) says that $d_j(\{b,b'\})=\langle f(\{a,b,b'\}):a\in E_j\rangle$, a coloring of $[B_j]^2$ with at most $2^{|E_j|}=2^{\mu(j-1)}$ colors, is constant."; Next investigation, "A positive outcome, with the block statement for (P2), proves the conjecture under $2^{\aleph_0}<\aleph_\omega$ by the endgame lemma."; lead page, The consequential gap and Next step 1, "admits in every full-size subset of $A_j$ a set $S_j\cup T_j$ on which the induced pair coloring is constant. A positive outcome, with the block statement behind (P2), proves the conjecture under $2^{\aleph_0}<\aleph_\omega$ by the endgame lemma." Defect: (P3) is the conjunction of the constancy of $d_j$ on $[B_j]^2$ and the constancy of its value on each lower $B_i$; the sentence keeps only the first, and the conditional conclusion has no bridge for the second. Witness: the coloring of W1 with $f(\{a,b,b'\})=\varphi(a)$ on every triple of pattern $(i,j,j)$, $\varphi$ not constant on $B_i$: $d_j$ is constant for every $B_j$ and (P3) fails. Replacement: state (P3) in (C) for $a\in S_i$ (the endgame lemma uses it only there), and record that a positive outcome gives (P3) so restricted only after thinning each $S_i$ to a fiber of $a\mapsto\langle v_j(a):j>i\rangle$, at most $2^{\aleph_0}$ values, which has size $\mu(i)$ under $2^{\aleph_0}<\mu(i)$ and is taken jointly with the (P1) fiber; or else pose the (P3) question with the value of $d_j$ required to be constant on each lower $B_i$, which is (P3) itself as a block statement. Either way, say which.
• F2 (suggested). Location: connections page, Next investigation, "a negative outcome, a coloring $f$ defeating (P3) in every choice of $S_j\cup T_j$, shows that a proof of the conjecture cannot pass through canonization over the lower blocks"; lead page, Next step 1, "A negative outcome shows that no proof passes through canonization over the lower blocks". Defect: the negative outcome refutes the realizability of the property carrying (P3), hence (C) for that $f$ and those blocks; it says nothing about other canonical forms, decompositions or parameter sets. Witness: none needed; the claim is a universal statement about proofs with no argument. Replacement: "shows that (C) fails for this $f$ on these blocks, so the route as formulated is closed; whether a weaker canonical form or another decomposition serves is not decided".
• F3 (suggested). Location: connections page, (P2) bullet, "inside which the relation $(2^\mu)^+\to(\mu^+,4)^3$ cannot be applied again ($|C'|$ is not of that form)". Defect: false when $\mu(i)=2^\nu$ for an infinite $\nu$, a case the sentence's own hypothesis $2^{\aleph_0}\le\mu(i)$ allows. Witness: $2^{\aleph_0}=\aleph_2$, $n(0)=1$, $2^{\aleph_2}>2^{\aleph_1}$, $\mu(0)=\aleph_2$, $|C'|=\aleph_3=(2^{\aleph_0})^+$. Replacement: "inside which the relation $(2^\mu)^+\to(\mu^+,4)^3$ applies only when $\mu(i)=2^\nu$, and then yields a set of color $0$ of size $\nu^+<\mu(i)$, since $\nu<\aleph_{n(0)}<\mu(i)$".
• F4 (note). Location: connections page, (P1) bullet, "Clause (3)'s heredity hypothesis is met by thinning $S_i$ alone and keeping $T_i$"; lead page, "clause (3) removes the remaining dependence when its hypothesis $2^{\chi+\kappa}<\operatorname{cf}\mu(i)$ holds". Defect: the property is not hereditary under same-size subsets in the reconstruction's reading, so clause (3) as stated does not apply; what is used, soundly, is the reconstruction's fiber argument with the thinning confined to $S_i$ (W2). Replacement: "the fiber argument of clause (3) applies with the thinning confined to $S_i$, which keeps $P_i$; the heredity hypothesis of (3) as stated is not needed".
• F5 (note). Location: connections page, after the endgame proof, "it shows that the case $\eta_2=1$, $\varepsilon=0$ cannot be handled without the $T_i$". Defect: $f_0$ violates hypothesis (ii) of the endgame lemma, so it defeats no implication the section states. Replacement: "it shows that the canonical values on the $S_i$ alone do not decide that case: $f_0$ realizes $\varepsilon=\eta_1=0$, $\eta_2=1$ with no color-$1$ four-element set and no color-$0$ set of the form $\bigcup S_i$".
• F6 (note). Location: connections page, (P2) bullet, "By (1A) at $j$ with $\bar a=(a,a')$ and the first thinning of the lemma page". Defect: the first thinning is a proof-internal fact of the reconstruction, outside the interface (T1) lists; it is essential here, because $P_i$ must be phrased with the points $a_j$ available at stage $i$. Replacement: add to (T1) the fact that every $b\in B_\alpha$ has the type of $a^*_\alpha$ over $E_\alpha$, or cite the reconstruction's labeled fact by name.
• F7 (note). Location: connections page, (P2) bullet, "So (P2) needs a block statement of its own: in every $C\subseteq A_i$ of size $\lambda_i$, for every coloring of $[C]^2$ with at most $2^{\aleph_0}$ colors"; Next investigation and lead Next step 1, the (P3) question "induced by an $f$". Defect: (BS2) quantifies over all colorings, stronger than the route needs, which is the one coloring $c_i$ determined by $f$ and the points $a^*_j$; and the (P3) question depends on the admissible lower sequence through $E_j$, not on $f$ and $C$ alone. Replacement: "is supplied by" for "needs", and "for every admissible lower sequence and every full-size $C$" in the (P3) question.
• F8 (note). Location: connections page, Outcome, "the lemma supplies (P1) under $2^{\aleph_0}<\aleph_\omega$". Defect: the realizability (H) behind (P1) also consumes the unheld relation $(2^\mu)^+\to(\mu^+,4)^3$, carried elsewhere on both pages but not in this sentence. Replacement: "supplies (P1) under $2^{\aleph_0}<\aleph_\omega$ and the imported triple relation".

Verdict

• S1: not a claim (the target). S2: not a claim (a definition).
• S3, the endgame lemma: refutation-failed.
• S4, the remark on the triples $T_i$: refutation-failed, with the scope of (c) narrowed as in F5.
• S5, the application of (T1) with the properties $P_i$: refutation-failed, conditionally on the imported relation $(2^\mu)^+\to(\mu^+,4)^3$, which is not held.
• S6, (P1) under $2^{\aleph_0}<\aleph_\omega$: refutation-failed, with the wording of F4.
• S7, the reduction of (P2) to the constancy of $c_i$: refutation-failed.
• S8a, the parenthetical "cannot be applied again": refuted-as-stated; its consequence S8b stands.
• S8b, the rest of the Erdős--Rado discussion and the need for (BS2): refutation-failed.
• S9a, no clause of the lemma reaches (P3): refutation-failed.
• S9b, "(P3) says that $d_j$ is constant": refuted-as-stated, witness in W1.
• S10, the counting obstruction for (P3): refutation-failed.
• S11, the falsification checks: refutation-failed.
• S12a, positive outcome plus (BS2) proves the conjecture under $2^{\aleph_0}<\aleph_\omega$: proof defective; statement unresolved. The bridge is missing (F1); it is supplied for (P3) restricted to $a\in S_i$ by the extra fiber step of W1, after which the corrected statement follows.
• S12b, a negative outcome excludes every proof through canonization over the lower blocks: proof defective; statement unresolved (F2).
• S13, the outcome sentence: refutation-failed as author-recorded, with the qualification of F8.

The lead page inherits these verdicts at the sentences that restate S3, S6, S7, S9b, S10 and S12; its provisional-premises list is accurate, and its standing sentences claim nothing beyond author-recorded.

Limitations. The relation $(2^\mu)^+\to(\mu^+,4)^3$ and the Erdős--Rado and Ramsey relations are imported, not held, and were not verified beyond their statements; the attribution of the triple relation was not checked. The sentences that cite the problem page's dated search and assessment were not checked, as that text is excluded. The proofs of (T1), (T2), (T3) and of the reconstruction pages are other reviewers' subjects and were used only through their stated interfaces.

This focused review assigns no tier and changes no status.