{"id":"e0ad1e87-b6ae-434f-b348-8f5e37865926","arxiv_id":"2608.22870","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Every 2-connected apex cubic graph is three-edge-colorable, and this completes the proof of Tutte's three-edge-coloring conjecture.","lead":"This paper proves that every 2-connected apex cubic graph admits a three-edge-coloring, closing the final open case of Tutte's three-edge-coloring conjecture from 1966. The proof extends the Four Color Theorem framework with large computer-checked discharging and reducibility steps and yields an O(n^2) coloring algorithm.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem depends on unverified computational core: Lemmas B.1–B.3 are outputs of custom code with no pinned commit, formal certificate, or exact pseudocode match for the cartwheel search.","rationale":"The reader's weakest assumption correctly identifies the load-bearing premise: the correctness and completeness of Lemmas B.1, B.2, and B.3. I agree with this assessment. My independent review of the manuscript found no specific mathematical error in the handwritten parts of the proof; the discharging identities, the connectivity reductions, and the multi-boundary island construction appear coherent. The concern is therefore not about the internal logic of the written argument but about the unverified computational core on which it rests. The manuscript itself contains admissions that strengthen this concern: Algorithm B.3.16 is not literally implemented (an unspecified shell script is used instead), and the refinement algorithm in Section 6.3.4 is deliberately tailored to the specific rule set with no general proof. Since the proof of Theorem 1.4—and hence the resolution of Tutte's three-edge-coloring conjecture—depends on these computations, a single bug would be fatal. The paper provides no formal machine-checked certificate, no pinned commit hash, and the AI reimplementations are explicitly not part of the mathematical justification. Thus the correct verdict remains CONDITIONAL: the theorem is plausible and the written proof is well-structured, but the computational lemmas need independent, reproducible verification before the claim can be accepted as established.","tokens_in":64768,"tokens_out":23308,"duration_ms":231077,"concrete_test":"Extract from the authors' code a machine-readable certificate for each of Lemmas B.1, B.2, and B.3: for B.1, the set of unblocked combined rules with their charges; for B.2, the complete list of cartwheels in C_all after pruning (which must be empty); for B.3, for each island in I, the maximal semi-consistent set of non-extendable ring colorings (which must be empty). Independently re-verify all three certificates with a separate simple checker not derived from the authors' code, and require a pinned commit hash for the source and data files. If any certificate fails or cannot be regenerated from the pinned source, the computational core is not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (Theorem 1.4) is logically sound only if the three computer-assisted lemmas B.1, B.2, and B.3 are complete and correctly executed. Lemma B.1 (Section 6.2) certifies that every unblocked free combination of discharging rules has charge at most 5, which is exactly Lemma 6.5. Lemma B.2 (Section 6.3.3) certifies that the cartwheel search for degrees 7–11 produces no counterexample, i.e., C_all = ∅, which is Lemma 6.6. Lemma B.3 (Section 7.1) certifies that every multi-boundary island in the generated set I is semi-D- or semi-C-reducible, which is Lemma 7.5. None of these lemmas has a hand proof; each is an assertion about the output of custom software. The manuscript gives pseudocode and a GitHub URL but no pinned commit hash, no machine-readable certificates, and no formalization of the algorithms. Moreover, the text following Algorithm B.3.16 states that the authors 'do not have source code that implements Algorithm B.3.16 literally' and instead used an unspecified shell script to loop over cartwheels. This breaks the pseudocode-to-code correspondence for the key degree-7-to-11 check. The AI reimplementations are explicitly disclaimed as 'not part of the mathematical justification.' Hence a single implementation bug—in the free-homomorphism routine, the pruning, the reducibility checker's handling of half-chains, or the enumeration of cartwheels—would invalidate Theorem 6.7 or Theorem 7.1 even if every handwritten lemma is correct.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves Theorem 1.4: every 2-connected apex cubic graph is three-edge-colorable. Since Robertson–Seymour–Thomas (Theorem 1.3) reduce Tutte's three-edge-coloring conjecture for all 2-connected cubic graphs without a Petersen minor to the doublecross and apex cases, and the doublecross case was already established by Edwards–Sanders–Seymour–Thomas, the paper claims to complete the proof of Tutte's conjecture. The proof follows the RSST97 paradigm: a minimal counterexample is reduced to a planar subcubic graph G whose dual G* is a triangulation with two or three digons; a set K of 915 normal configurations is used with a discharging system consisting of the 32 rules of RSST97 plus three new digon-handling rules. The discharging argument establishes that some configuration in K appears in G* (Theorem 6.7), and the reducibility argument establishes that this configuration yields a semi-reducible multi-boundary island in G (Theorem 7.1). Three computational lemmas are load-bearing: Lemma B.1 (free discharging combinations not blocked by K transfer charge at most 5), Lemma B.2 (the degree 7–11 cartwheel search finds no bad local structure), and Lemma B.3 (every island in the generated set I is semi-D- or semi-C-reducible). The appendix contains detailed pseudocode, and the authors provide GitHub code and independent AI-based reimplementations, explicitly stating that the AI reconstructions are not part of the mathematical justification.","tokens_in":65119,"tokens_out":3336,"duration_ms":37165,"significance":"If Theorem 1.4 is correct, this is a landmark result: it completes Tutte's three-edge-coloring conjecture and, with it, the long-standing classification of 3-edge-colorability for 2-connected cubic graphs without a Petersen minor. The proof framework is a genuine generalization of the RSST97 discharging and reducibility program to graphs with digons, and the manuscript is unusually transparent about its computational dependencies: it provides detailed pseudocode, source-code links, an explicit list of the auxiliary rules, and even reports an independent reproducibility check by generative-AI systems. The mathematical narrative is coherent, the structural lemmas are carefully written, and the proof does not appear to assume the target theorem; it relies instead on the Four Color Theorem, the RST97 reduction, ESST16's doublecross theorem, and the authors' own previously published algorithms. The main weakness is that the three computational lemmas are certified only by running custom software, with no pinned commit, no machine-readable certificates, and no formal proof of pseudocode-to-code correspondence.","major_comments":[{"comment":"The claim that every free combination of discharging rules not blocked by K transfers charge at most 5 (Lemma B.1, supporting Lemma 6.5) is an assertion about the output of custom software. The manuscript gives pseudocode and a GitHub URL but no pinned commit hash, no machine-readable certificate, and no formalization of the algorithms. A single bug in the free-homomorphism routine, the pruning of combined rules, or the enforcement of single digon incidence would invalidate Lemma 6.5 and hence the discharging proof of Theorem 6.7 even if all handwritten lemmas are correct. I recommend that the authors provide a version-pinned source snapshot with output hashes, and ideally machine-checkable certificates or a formally verified implementation of Lemmas 6.9–6.12 and the final combinatorial check.","section":"Section 6.2 / Lemma B.1"},{"comment":"The degree 7–11 cartwheel search is certified by Lemma B.2, which is exactly the assertion that the assertion in Algorithm B.3.15 does not fail in every call. Yet the manuscript states, immediately after Algorithm B.3.16, that the authors 'do not have source code that implements Algorithm B.3.16 literally' and instead used a shell script to loop over cartwheels. This breaks the pseudocode-to-code correspondence for the key check that establishes Lemma 6.17 and hence Lemma 6.6. The authors should either provide a literal implementation of Algorithm B.3.16, give a formal argument that the shell-script loop is equivalent to lines 3–5, or supply a hand-checkable certificate of the cartwheel search. Without this, Lemma 6.6 is not independently verifiable from the manuscript.","section":"Section 6.3.3 / text after Algorithm B.3.16"},{"comment":"Lemma 7.5, which asserts that every multi-boundary island in I is semi-D- or semi-C-reducible, is backed only by the software described in Appendix B.4 and Lemma B.3. The reducibility checker must handle half-chains, all ring colorings beyond the usual parity-restricted set, and the full enumeration of deletable edge sets; these are exactly the places where an implementation error could go unnoticed. Because Theorem 7.1 and hence Theorem 1.4 depend on Lemma 7.5, the manuscript should include a machine-checkable certificate from the checker (for example, a witness semi-consistent set for each island, or a formally verified checker), or an independent deterministic reimplementation that is included in the mathematical justification rather than merely as a reproducibility aid.","section":"Section 7.1 / Lemma B.3"}],"minor_comments":[{"comment":"The GitHub repositories should be described with pinned commit hashes, the software environment and dependencies, and SHA-256 or similar hashes of the generated data files and outputs, so that the reader can reproduce the exact results claimed in Lemmas B.1–B.3.","section":"General / GitHub reproducibility"},{"comment":"In the sentence 'since G has two 6 or three vertices of degree two', the '6' appears to be a typographical artifact; the intended statement is presumably 'two or three vertices of degree two'.","section":"Section 1.1"},{"comment":"The notation R for the rule set and R_e for the set of rules applied to a dart e is clear in context, but the simultaneous use of R for a single rule and for the rule set in Definition 6.1 and Lemma 6.5 may confuse readers; a distinct symbol for the rule set, such as Script-R, would improve readability.","section":"Section 6.2"},{"comment":"The text says 'We add a footnote in this part of the algorithm,' but no footnote is present in the arXiv text; either include the footnote or remove the parenthetical reference to it.","section":"Section 7.1 / Algorithm B.4.1"},{"comment":"The complexity expression 'Opn 2q' should be typeset as O(n^2) to match standard notation; this is a purely presentational issue.","section":"Section 8 / formatting"}],"recommendation":"major_revision","confidential_remarks":"This is a very strong paper whose main theorem, if correct, closes a major open problem. My recommendation is major_revision rather than reject because the mathematical framework appears coherent and the computational gaps are, in principle, fixable within the paper's scope: the authors can pin the code, provide certificates, or close the shell-script gap. However, the current manuscript cannot be accepted as-is in a rigorous journal because the three computational lemmas are load-bearing and the paper itself admits that no literal implementation of Algorithm B.3.16 exists. I would also note that the independent AI reimplementations are explicitly disclaimed as not part of the mathematical justification, so they cannot substitute for a machine-checked certificate or a formal proof of equivalence. The editor may wish to consult an expert on computer-assisted proof verification concerning the standards expected for such lemmas."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing you need to know: this paper supplies the missing apex case, so if the proof is right it settles Tutte's three-edge-coloring conjecture. Theorem 1.4 is the last piece after RST97 and ESST16, and the authors earnestly extend that framework with multi-boundary islands, configurations with digons, and modified dart representations. The handwritten structural lemmas are detailed and internally coherent; I did not find a gap in Sections 2-5, and the discharging setup in Section 6 is explained carefully.\n\nCredit where it is due: the computational appendix is unusually honest. Pseudocode is given for the newly introduced routines, the code is on GitHub, and the authors explicitly say the AI reimplementations are not part of the mathematical justification. They also admit, right after Algorithm B.3.16, that they do not have source code implementing that algorithm literally and used a shell script instead. That kind of transparency is rare and helps the reader see exactly what is and is not machine-checked.\n\nThe soft spot is exactly what the stress-test note says, and it is real: Lemmas B.1-B.3 are load-bearing, and they are outputs of custom software with no pinned commit, no machine-readable certificate, and no formal proof object. The gap between Algorithm B.3.16 and the shell script matters because the degree-7-to-11 cartwheel search is the core of Lemma 6.6. A single bug in the free-homomorphism routine, the pruning, or the reducibility checker's half-chain handling would invalidate the discharging argument even if every handwritten lemma is correct. I do not find the flaw fatal, but the proof is conditional in a way that no amount of handwritten detail can patch.\n\nThe central argument, as far as I can tell, holds up. The reduction to three computer lemmas is clean, and the lemmas are stated precisely. The paper deserves a serious referee, not a desk reject. I would send it to peer review with the explicit instruction that the referee verify the code-pseudocode correspondence, ideally by running the authors' checks and the independent AI reconstructions. If that audit passes, this becomes a landmark result; if it fails, the error will be local but decisive.\n\nWho gets value: graph theorists working on edge-coloring, flows, and reducibility, plus algorithm designers who want the O(n^2) coloring procedure. I would cite it if the computational audit checks out.","headline":"If the computer checks hold up, this completes Tutte's three-edge-coloring conjecture; the proof is serious and the computational transparency is above average, but the cartwheel-search gap needs an independent audit.","tokens_in":778,"tokens_out":688,"would_cite":true,"duration_ms":26393,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"pith_extraction":null,"created_at":"2026-08-28T00:31:26.486766+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":null,"supporting_citations":[],"review_version":1}