---

Subject and independence

Role. The reviewer worked in a fresh context from the assignment alone and took no part in writing the connections page, the reconstruction pages, the library cards or the problem pages. The charge was refutation. This review assigns no tier and changes no status.

Subject. Page wiki/research/erdos_354/connections.md as it stood at 2026-09-28T13:50:48Z, section "## Problem 348: deletion thresholds", lines 490--516 (the H2 heading at line 490 through the "Outcome." paragraph at line 516; the next H2 opens at line 518). The assignment named no separate lead page; the connections page is the only subject page. Its shared preamble (lines 1--58: purpose and standing, the notation for $A_{\alpha,\beta}$, complete and strongly complete sets, the normalized pair, and the dyadic-rational and conjecture conventions) was read as input. The bodies of the page's other sections were not read; only the list of H2 headings was used to locate the section.

Read set and depth.

• The section, read in full and re-derived sentence by sentence; the preamble, read in full.
• Reconstruction pages, read in full as premises at their reviewed standing: Yu--Chen Theorem (statement, Steps 1--4, dependency map, scope), Geneson Theorem 9 and Geneson Corollary 12 (statements, proofs, scope). The folder's remaining reconstruction pages were identified by name only; the section uses nothing from them beyond the theorem's conclusion.
• Problem pages, Statement paragraphs only: Problem 348 and Problem 354.
• Library: the provenance and read-status paragraphs of the Geneson (2026) card and the provenance and Statement sections of its result pages Theorem 1, Theorem 9 and Corollary 12; the provenance and Statement sections of the Yu--Chen Theorem result page. The held Geneson PDF (arXiv:2609.25107v1, 14 pages) was read in its text layer, pp. 1--5 in full (Theorem 1; Section 2 with Theorem 2, Lemma 3, Lemma 4, Proposition 5 and the exclusion of $m\ge2$), together with the closing line of Section 4 (p. 11) and the roadmap paragraph of the introduction (p. 3). No page image was needed: the section quotes no formula from the source.
• Guidance: docs/verification.md sections "Grading and claim standing", "Audit checklist — the canonical failure modes", "Whole-claim report" and "Audit checklist"; docs/approach_vetting.md; docs/evidence.md section "Source fidelity"; docs/math_authoring.md.

Exposures. Three incidental exposures, none of review, grade or assessment text, disclosed in full:

1. A grep for headings and bold labels over the two problem pages, run to locate their Statement sections, printed each bold label line of those pages. For E0348 this included the line "Status. Open. Tags. number theory, complete sequences." (line 21); for E0354 it included the label lines of its header block (lines 38, 52, 97, 128, 139, 150, 215) and the opening words of the label lines of its Current assessment section (lines 243, 303, 351, 384, 404, 419), and nothing of their bodies. None of these lines concerns Problem 348's mathematics; the E0348 status word was not used except where the section itself reports it (S2).
2. A grep for the words "pending", "compilation task" and "reconstruct" over the four pages of the Geneson card folder, run to check the section's sentence about a pending task (S5b), printed one line of theorem_1.md (line 72), three lines of theorem_9.md (lines 57, 59, 80) and two lines of corollary_12.md (lines 65, 67), all outside those pages' Statement sections. Their content is reported under F1 and nowhere else.
3. Directory listings by name only: the folder wiki/research/erdos_354/ including the name of its evidence/ subtree and the file names under evidence/verify/. No file under evidence/ was opened.

The folder's _index.md, every existing review and grade, every Current assessment, Progress, Known results and standing text, and the web were not read.

Restatement

The section makes the following statements. The page's preamble fixes "complete" as: every sufficiently large integer is a sum of distinct elements, and "strongly complete" as: complete after every finite deletion.

• S1 (target). Problem 348 asks for which integer pairs $0\le m<n$ there exists a complete nondecreasing sequence of integers that remains complete after the removal of any $m$ elements and is not complete after the removal of any $n$ elements. ("Any" is universal in both clauses: every $m$-element removal keeps completeness; every $n$-element removal destroys it.)
• S2 (records). (a) The problem page records the status open with progress not compiled. (b) The Geneson card records Theorem 1, the classification "exactly the pairs with $m\le1$", as a preprint claim read for structure only. (c) The folder index names Problem 348 only as a further target of the same preprint.
• S3 (nothing transports). Three supporting statements.
  • S3a. The folder's reconstructions of Geneson's paper concern Theorem 9 and Corollary 12 and not Theorem 1.
  • S3b (the obstruction). The Yu--Chen theorem gives completeness after every finite deletion. Hence for every pair $\alpha,\beta>0$ satisfying the theorem's hypothesis ($\alpha/\beta$ irrational), the set $A_{\alpha,\beta}$ has Problem 348's first property for every $m\ge0$ and fails its second property for every $n\ge1$; so no such set is an example for any pair $(m,n)$.
  • S3c. The content of the problem beyond the two classical constructions is the exclusion of $m\ge2$; in Geneson's paper this is carried by an incompleteness mechanism, the central-interval Theorem 2 and Lemma 4; the folder holds no reconstruction of it; and the Yu--Chen descent argument has no counterpart to it.
• S4. The folder gives no part of the classification, neither the existence side ($m\le1$) nor the exclusion side ($m\ge2$).
• S5. (a) No next investigation arises from the folder. (b) A reconstruction of Geneson's Sections 2--4 is a separate compilation task that the card records as pending.
• S6. The transfer fails; the only link between the folder and Problem 348 is the shared source.

S3b is the section's one mathematical statement and its stated obstruction; S3a, S3c (first three clauses) and S5b are record and source-fidelity statements; the "no counterpart" clause of S3c, S5a and the first clause of S6 are judgments.

Checklist

• Quantifiers and scope. Pass. S3b needs the theorem's conclusion in its full universal form (every finite $F$), which the reconstruction states; the section's "for every $m$" and "for no $n$" follow from it without an almost-all step. The hypothesis $\alpha/\beta$ irrational is carried only by the phrase "of its theorem"; it is load-bearing (see Weakest steps, W3), so F2 suggests making it explicit. The pair constraint $n>m\ge0$ forces $n\ge1$, which the "for no $n$" clause needs.
• Circularity. Inapplicable. No statement assumes its own conclusion; S3b consumes the reconstructed theorem and the problem's definition.
• Model and convention changes. Pass. Problem 348 concerns a nondecreasing sequence with removals of elements; the theorem concerns the set $A_{\alpha,\beta}$ and sums of distinct elements. The bridge is checked under both readings (distinct values in increasing order; the interleaving with repeated occurrences) in W1 and holds under both. The completeness convention (eventual, distinct indices) is the same in the preamble, in the problem statement and in the source (its p. 1).
• Finite and statistical overreach. Inapplicable. No finite sample or heuristic is used.
• Uniformity. Inapplicable. The threshold $H$ of the theorem depends on $F$, and the section claims nothing uniform in $F$, $m$ or $n$.
• Extremal conclusions. Inapplicable. No infimum, supremum or sharpness sentence.
• Consequences and composition. Pass. The "so" of S3b is re-derived in W1. S3's conclusion "Nothing" composes S3a, S3b and S3c: the folder has no examples for any pair and no argument on the exclusion side; S4 and S6 restate this. The judgment clauses are recorded as judgments.
• Computation. Inapplicable. The section runs no computation.
• Reproduction. Inapplicable. No rerun command or coverage claim.
• Source and verdict fidelity. One failure and two imprecisions. The labels "central-interval Theorem 2", "Lemma 4", "the exclusion of $m\ge2$" and "Sections 2--4" match the held PDF (Theorem 2 is labeled "(central-interval)" on p. 3; Lemma 4 is the two-occurrence gap lemma on p. 4; "Exclusion of $m\ge2$ in Theorem 1" is on p. 5; the roadmap on p. 3 assigns Sections 2--4 to Theorem 1). S5b's clause "the card records as pending" is not supported by the card (F1, refuted-as-stated). S2b compresses the card's read status (F3, note). S3c's parenthetical names a representation theorem as part of the incompleteness mechanism (F4, note). S2a and S2c lie outside the read set and are unverified.

Weakest steps

W1: the bridge in S3b (the theorem's conclusion to Problem 348's two properties). Fix $\alpha,\beta>0$ with $\alpha/\beta$ irrational and write $A=A_{\alpha,\beta}$, an infinite set of positive integers. The reconstructed theorem says: for every finite $F\subseteq\mathbb Z$ there is $H$ such that every integer $m\ge H$ is a sum of distinct elements of $A\setminus F$.

Reading 1: the sequence is $A$ in increasing order (distinct values). Removing any $m$ elements means passing to $A\setminus F$ with $|F|=m$; the theorem gives completeness of $A\setminus F$; so the first property holds for every $m\ge0$ (with $F=\emptyset$ giving completeness of $A$ itself). The second property for a given $n$ says that every removal of $n$ elements leaves an incomplete set; the theorem says every such removal leaves a complete set; since $A$ is infinite there is at least one $n$-element removal, so the second property fails for every $n\ge1$. Pairs have $n>m\ge0$, hence $n\ge1$: no pair $(m,n)$ is witnessed.

Reading 2: the sequence is the interleaving of $(\lfloor2^k\alpha\rfloor)_{k\ge0}$ and $(\lfloor2^k\beta\rfloor)_{k\ge0}$ with repeated values kept as separate occurrences and sorted. Delete any $n$ occurrences. Let $F$ be the set of nonzero values all of whose occurrences were deleted; $|F|\le n$. Every element of $A\setminus F$ still has an occurrence in the remaining multiset, and distinct values have distinct positions, so every sum of distinct elements of $A\setminus F$ is a sum of remaining occurrences at distinct positions. By the theorem the remaining multiset is complete. Zero occurrences (when $\lfloor\alpha\rfloor=0$ or $\lfloor\beta\rfloor=0$) change no sum and their deletion changes no sum. So both properties behave exactly as in Reading 1. The bridge holds under either convention, and no zero, index or multiplicity subtlety breaks it.

W2: the characterization in S3c of the exclusion of $m\ge2$. From the held PDF, Section 2: Lemma 3 (single-deletion tolerance forces the prefix slack $S_j-b_{j+1}$ to infinity, also after any fixed finite deletion); Theorem 2 (a complete nondecreasing positive sequence with threshold $T$ and diverging slack represents all of $[T,S_N-T]$ from its first $N$ terms for large $N$); Lemma 4 (a deletion-minimal sequence with such central intervals has, for every $C$, two occurrences whose deletion leaves arbitrarily late runs of more than $C$ consecutive missing integers); Proposition 5 (if every two-occurrence deletion keeps completeness, every finite deletion $D$ that keeps completeness can be enlarged by one occurrence keeping completeness: otherwise $A\setminus D$ is deletion-minimal, Lemma 3 and Theorem 2 supply its central intervals, Lemma 4 with $C=\sum D$ supplies two occurrences $E$ whose deletion leaves gaps that restoring $D$ cannot fill, so $A\setminus E$ is incomplete, a contradiction). The exclusion then reads: an $m$-tolerant sequence with $m\ge2$ is two-tolerant, so some $n$-deletion keeps completeness, against the required universal failure; the nonpositive-term and bounded cases are reduced separately (p. 5). The engine that produces the contradiction is Lemma 4's construction of long gaps after two deletions, an incompleteness statement; Theorem 2 is a representation statement that supplies Lemma 4's hypothesis. The section's "incompleteness mechanism" describes the engine correctly; its parenthetical folds the input theorem into the mechanism (F4). The clause "the folder holds no reconstruction" is confirmed by the folder's page names: the two Geneson reconstruction pages treat Theorem 9 and Corollary 12 only, and no page treats Theorem 2, Lemma 4 or Proposition 5.

W3: the scope phrase "of its theorem" in S3b. Without the restriction to the theorem's hypothesis the sentence would be false. Witness: $\alpha=1$, $\beta=2$ give $A_{1,2}=\{1,2,4,8,\dots\}$, the powers of two (the two floor sequences coincide up to a shift, and $0$ is excluded by definition). This set is complete (binary expansion) and, after the removal of any $n\ge1$ elements, one of them $2^k$, every sum of distinct remaining elements has binary digit $k$ equal to $0$, so the infinitely many integers with that digit equal to $1$ are missed: it is an example for every pair $(0,n)$. Here $\alpha/\beta=1/2$ is a power of $2$, outside the theorem's hypothesis and outside the rational-ratio lines the page's first section transports (its conjecture convention excludes power-of-two ratios). So the phrase "of its theorem" carries the irrational-ratio hypothesis and is load-bearing; the sentence is correct as written, and F2 asks only that the hypothesis be named.

Strongest attack

• S1. Compared clause by clause with the Statement paragraph of the problem page ("remains complete after removing any $m$ elements, but $A$ is not complete after removing any $n$ elements", for $0\le m<n$ and a nondecreasing sequence of integers). The paraphrase drops "of integers" and the nondecreasing ordering, neither of which changes the question for the section's purpose. Failed.
• S2. (b) Attacked by comparison with the card's read-status paragraph and the Theorem 1 result page: both record that the statement of Theorem 1 was read clause by clause and that its proof was read for structure only. The section's "a preprint claim read for structure only" attaches the structural reading to the claim rather than to its proof; it understates the card and changes no standing. Failed, with F3. (a) and (c) could not be attacked within the read set.
• S3a. Attacked by listing the folder's pages: the Geneson reconstructions are the Theorem 9 and Corollary 12 pages, and the Corollary 12 page rests on Theorem 9 and Proposition 11 only. Failed.
• S3b. Four attacks. (i) Convention: set against multiset; W1 shows the bridge holds under both. (ii) Quantifier: "not complete after removing any $n$ elements" read existentially ("some $n$-removal leaves it incomplete") would still fail for the theorem's sets, since every removal leaves them complete; read universally it fails a fortiori. No reading rescues a pair. (iii) Zero and multiplicity: $0$ is excluded from $A_{\alpha,\beta}$ and zero occurrences change no sum. (iv) Hypothesis: the powers-of-two witness of W3 shows the sentence would be false without the restriction to the theorem's sets, and the sentence carries that restriction. Failed.
• S3c. Attacked by arguing that the exclusion is a completeness mechanism, since Proposition 5's conclusion is completeness after some larger deletion. The conclusion is of that form, but its proof is by contradiction and the step doing the work is Lemma 4's production of arbitrarily long gaps; the characterization of the engine stands, and the parenthetical is imprecise only in naming Theorem 2 as part of it (F4). The "no counterpart" clause is a judgment about the descent argument, whose conclusion (a decreasing modular-run invariant forcing representation) points the opposite way from a deletion that destroys completeness; no counterexample to a judgment is possible, and none is claimed. Failed.
• S4. Follows from S3; no separate attack surface. Failed.
• S5. (b) Attacked by reading the card's provenance and read-status paragraphs, which record the structural reading of Sections 2--4 and no task, and by a word search over the card folder (exposure 2), which found the word "pending" only in sentences on the Theorem 1 and Theorem 9 result pages saying that a separate triage of a problem page is pending. The card records no pending reconstruction or compilation task. Succeeded: refuted-as-stated, F1. (a) is a judgment.
• S6. The clause "the only link is the shared source" was checked against S3a--S3c: both Theorem 1 and the reconstructed Theorem 9 and Corollary 12 come from the same preprint, and nothing else in the read set connects the folder to the problem. Consistent; the rest is a judgment.

Premises

• Yu--Chen Theorem reconstruction (page), read in full. Interface used: for $\alpha,\beta>0$ with $\alpha/\beta$ irrational and every finite $F\subseteq\mathbb Z$ there is $H$ such that every integer $m\ge H$ is a sum of distinct elements of $A_{\alpha,\beta}\setminus F$; in particular $A_{\alpha,\beta}$ is strongly complete. Standing: an author-recorded reconstruction of an unrefereed manuscript, independently reviewed at the standing the assignment states; used here as a premise at that standing. The section uses only this conclusion, within its hypothesis.
• Geneson Theorem 9 and Corollary 12 reconstructions (Theorem 9, Corollary 12), read in full. Used only to confirm S3a: their subjects are Theorem 9, Proposition 11 and Corollary 12 of the source; neither touches Theorem 1, Theorem 2, Lemma 3, Lemma 4 or Proposition 5.
• Geneson (2026), arXiv:2609.25107v1, held PDF (card), text layer, pp. 1--5 read in full and p. 11 (closing line) and p. 3 (roadmap) located. Interfaces: Theorem 1 (p. 1); Theorem 2 (p. 3), Lemma 3 (p. 3), Lemma 4 (p. 4), Proposition 5 (p. 4), the exclusion of $m\ge2$ (p. 5), as restated in W2. Card read status: statement of Theorem 1 checked clause by clause; proof read for structure only. The result pages' Statement sections for Theorem 1, Theorem 9 and Corollary 12 were read. A preprint; nothing in it is used as established, only as the object the section characterizes.
• Yu and Chen (2026), manuscript of 13 September 2026, result page (Theorem), provenance and Statement read; the PDF was not opened, since the section invokes the theorem through the reconstruction.
• Problem statements. E0348 and E0354 Statement paragraphs, read in full; the section's S1 is checked against E0348.
• Explicit assumptions. The preamble's definitions of complete and strongly complete; the source's convention that completeness is eventual and counts occurrences. No batch acceptance order.

Findings

• F1 (required). Location: lines 512--514, "A reconstruction of Geneson's Sections 2--4 is the separate compilation task the card records as pending." Defect: source fidelity. The card records that the proof of Theorem 1 (Sections 2--4) "was read for structure only"; it records no reconstruction or compilation task and does not use the word "pending" for one. Witness: the card's read-status paragraph (lines 50--61 of its _index.md); the word search of exposure 2, whose only hits for "pending" are the result pages' sentences that a separate triage of a problem page is pending. Proposed replacement: "A reconstruction of Geneson's Sections 2--4 would be a separate compilation task; the card records that this proof was read for structure only."
• F2 (suggested). Location: lines 501--504, "the Yu--Chen mechanism produces completeness after every finite deletion: the sets $A_{\alpha,\beta}$ of its theorem satisfy the first property for every $m$ and the second for no $n$". Defect: the irrational-ratio hypothesis is carried only by "of its theorem", and it is load-bearing: the powers-of-two witness $A_{1,2}$ of W3 is an example for every pair $(0,n)$. Proposed replacement: "the sets $A_{\alpha,\beta}$ of its theorem ($\alpha/\beta$ irrational) satisfy ...".
• F3 (note). Location: lines 495--496, "the Geneson card records Theorem 1, the classification $m\le1$, as a preprint claim read for structure only." Defect: the card records the statement as read clause by clause and the proof as read for structure only; the sentence attaches the structural reading to the claim. It understates and changes no standing. Proposed replacement: "as a preprint claim whose proof was read for structure only."
• F4 (note). Location: lines 505--506, "an incompleteness mechanism (Geneson's central-interval Theorem 2 and his Lemma 4)". Defect: Theorem 2 is a representation theorem (central intervals are represented); the incompleteness step is Lemma 4, reached through Proposition 5, with Lemma 3 and Theorem 2 as its inputs (W2). Proposed replacement: "an incompleteness mechanism (Geneson's Lemma 4, fed by Lemma 3 and the central-interval Theorem 2 through Proposition 5)".

Verdict

• S1: refutation-failed. A faithful paraphrase of the problem statement.
• S2: refutation-failed for clause (b), with the imprecision of F3; clauses (a) and (c) lie outside the commissioned read set and are unverified.
• S3a: refutation-failed.
• S3b: refutation-failed. The bridge from the theorem's universal conclusion to both of Problem 348's properties holds under both sequence conventions; the restriction to the theorem's sets is necessary and present (F2).
• S3c: refutation-failed for the record clauses; the parenthetical is imprecise (F4); the "no counterpart" clause is a judgment, not a claim.
• S4: refutation-failed, as a consequence of S3.
• S5: clause (a) is a judgment, not a claim; clause (b) refuted-as-stated (F1): the card does not record the task the sentence attributes to it.
• S6: the "shared source" clause is consistent with S3a--S3c; the rest is a judgment, not a claim.

Limitations. The review covers the section's sentences and the preamble's conventions; the proofs of the reconstruction pages were taken as premises at their reviewed standing and not re-derived here. The source PDF was read in its text layer only, which suffices because the section quotes no formula from it. S2(a) and S2(c) refer to text outside the read set and were not checked. No computation was run; the powers-of-two witness of W3 is a two-line argument.

This focused review assigns no tier and changes no status.