{"id":"ea7f2c30-f738-428b-9dfd-a1566f354863","arxiv_id":"2607.05835","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Danus constructs an integral matroid tangent class in K_Z(M,G) that matches wonderful-model tangents, recovers Chow Hilbert series via HRR, and obeys Chern-alpha bounds, reproducing Cheng's concurrent work.","lead":"An AI agent autonomously built an integral K-theory tangent class for every loopless matroid and building set, matching the tangent bundle of wonderful compactifications when realizable. The write-up is a faithful dump of that agent run, plus a concurrent human paper it independently rediscovered.","discovery_kind":"replication","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified beyond the already-flagged incomplete write-up of Lemma 8.7.","rationale":"The paper's central claim is the existence of an integral class T^Z_{M,G} with three properties. The construction and the first two properties are supported by a coherent chain (integral descent via saturated ϕ_K, linkage Prop. 4.3, realizable θ-descent, one-flat Hilbert recursion anchored at the maximal model). The only incomplete local justification is exactly the one the reader flagged; it is confined to the Chern-alpha induction for ranks ≥5 and is already acknowledged by the authors as non-critical for the rest of the theorem. Because that gap is already reflected in the CONDITIONAL verdict and no stronger load-bearing concern appears, the stress-test does not move the verdict. The AI-origin of the text raises readability and compression issues but does not, on the written argument, introduce an additional correctness risk beyond the documented lemma.","tokens_in":37501,"tokens_out":633,"duration_ms":7053,"concrete_test":"Insert the short proof of the truncation identity from Che26 Prop. 4.19 (or an independent derivation of the binomial transfer (8.1)) into the manuscript at Lemma 8.7; re-check that Prop. 8.9 and Thm. 8.15 then close without external citation. If the identity holds and the induction goes through, clause (iii) is fully self-contained and the CONDITIONAL status can be lifted for the mathematical claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reader's weakest_assumption correctly isolates the only soft spot that touches a claimed property: Lemma 8.7's binomial-transfer paragraph is overcompressed (authors' own remark, §8), so the inductive step of the nested-Segre-tail formula (Prop. 8.9) for k≥4 is not self-contained. That step is used only for the Chern-alpha lower bound (Thm. 8.15 / clause (iii) of Thm. 1.1). The construction of T^Z itself (Props. 4.2–4.3), the integral realizable comparison (Prop. 6.5 + §6), and the Hilbert identity (Thm. 7.8 via Prop. 9.1) never invoke the truncation transfer; they rest on saturated one-step descent (Prop. 5.5), BEST input, and the one-flat recursions of §§7–8 that are independent of Lemma 8.7. Because the authors already record that the lemma is true by a short external argument (Che26 Prop. 4.19) and that the gap does not affect construction or P^K=Hilb, the incompleteness is a local expositional defect rather than a load-bearing crack in the central claim. No further internal inconsistency or hidden assumption was found that would undermine the existence of the integral class or clauses (i)–(ii).","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper constructs, for every loopless matroid M and every top-containing Feichtner–Yuzvinsky building set G, an integral class T^Z_{M,G} = sum_{F in G^circ} (1-τ_F)^{-1} - Q^Z_G in the combinatorial K-ring K_Z(M,G). The integral quotient Q^Z_G is obtained by saturated one-step descent of a rank-corrected BEST tautological quotient from the maximal model (Props. 4.1–4.2, 5.5). After rationalization the class recovers the rational tangent class of §3, and the three claimed properties are established: (i) under a realization L there is an integral unital isomorphism θ^Z_G : K_Z(M,G) → K_0(W_{L,G}) sending T^Z to [T_{W_{L,G}}] (Prop. 6.5 and §6); (ii) the K-theoretic Todd polynomial of the rationalization equals the Chow Hilbert series (Thm. 7.8 via Prop. 9.1); (iii) the Chern-alpha numbers satisfy the binomial lower bounds (Thm. 8.15 via Prop. 9.3). The construction and proofs occupy §§2–11; the bulk of the text is presented as the autonomous output of the agent Danus, with human editorial remarks and an experimental log in Appendix B.","tokens_in":37890,"tokens_out":1106,"duration_ms":10873,"significance":"If correct, the result supplies a uniform integral K-class for arbitrary (not necessarily realizable) matroids and building sets that specializes to the geometric tangent bundle on wonderful models, recovers the Hilbert series by HRR, and obeys the expected Chern-alpha inequalities. The integral lift rests on an explicit free-cokernel/saturation argument for one-step maps (Prop. 5.5) and an integral θ-descent (Prop. 6.5), both of independent interest for combinatorial K-theory. The concurrent reproduction of the first author’s arXiv:2606.22650, obtained without access to that preprint, is a noteworthy demonstration of agentic mathematical search; the fact-graph statistics and verification pipeline documented in Appendix B give a concrete, inspectable record of the experiment.","major_comments":[{"comment":"§8, Lemma 8.7 (rank-(k+1) truncation transfer) is used as the inductive step for the nested-Segre-tail formula (Prop. 8.9) when 4 ≤ k ≤ r−2, and therefore for the Chern-alpha lower bound (Thm. 8.15 / clause (iii) of Thm. 1.1). The authors themselves flag that the binomial-transfer paragraph is incomplete as written and that a short external proof appears in Che26, Prop. 4.19. The construction of T^Z, the realizable comparison, and the Hilbert identity never invoke this lemma, so the gap is local to clause (iii). A self-contained write-up of the truncation identity (or an explicit citation of the short argument) is required before the Chern-alpha claim can be regarded as fully internal to the manuscript.","section":null}],"minor_comments":[{"comment":"Several editorial remarks (e.g., after Prop. 3.3 Step 2, after Lemma 8.7, after Prop. 8.6) correctly flag compressed passages; expanding those few lines would improve readability without changing the logic.","section":null},{"comment":"Notation for the rational versus integral classes (T vs T^Z, Q vs Q^Z) is introduced late relative to the prompt’s abuse of notation; a short notational table at the start of §2 would help.","section":null},{"comment":"Appendix B’s fact-graph statistics are valuable but dense; a one-paragraph summary of how many facts support each of the three clauses of Thm. 1.1 would orient the reader.","section":null},{"comment":"The fan-support guard (Thm. 10.1) and the uniform atom-plus-top example (Cor. 10.2) are useful; a single sentence in the introduction pointing to them would clarify that the theorem is intrinsic and does not require completeness of the nested fan.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The manuscript is simultaneously a research paper and a carefully documented AI-experiment report. The mathematical core (construction + clauses (i)–(ii)) appears sound; the only load-bearing expositional gap is the incomplete write-up of Lemma 8.7, already acknowledged by the authors and confined to the Chern-alpha bound. I recommend minor revision rather than major revision because the fix is local and the external short proof is already cited. The concurrent self-citation of Che26 is fully disclosed and does not create a circularity problem. Scope is appropriate for a specialized algebraic-geometry / combinatorial-geometry journal; the AI-experiment material is of independent interest but should not obscure the mathematical evaluation."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The punchline is that this paper constructs a genuine integral tangent class T^Z in K_Z(M,G) for every loopless matroid and top-containing Feichtner–Yuzvinsky building set, with the three expected properties (realizable specialization to the wonderful-model tangent bundle, P^K = Hilb via HRR, and Chern-alpha lower bounds). It is explicitly a reproduction of the first author's concurrent Che26, obtained independently by the Danus agent before that preprint was public, and the authors document the run carefully in Appendix B.\n\nWhat is new is the AI-derived path and the detailed fact-graph experiment, not the mathematical object itself. The math that is here is done carefully: the rational backbone (BEST descent of Newton sums, reconstruction via filtered Chern character, one-flat Hilbert recursion) is written out, the integral lift uses saturated one-step maps with free cokernel (Prop. 5.5) and an explicit generator formula, and the realizable theta descent is integral before rationalization. Construction precedes the properties, so circularity is low. The fan-support guard is honest about non-complete nested fans.\n\nThe soft spot is exactly the one the authors flag: Lemma 8.7 (rank-(k+1) truncation transfer) has an overcompressed binomial-transfer paragraph, so the inductive step of the nested-Segre-tail formula for k ≥ 4 is not self-contained. That lemma is used only for the Chern-alpha bound; construction, realizable comparison, and the Hilbert identity never invoke it. The authors note the lemma is true by a short argument in Che26 Prop. 4.19 and that the gap does not affect the main claims. That is a local expositional defect, not a load-bearing crack.\n\nThis is for people already working in matroid Chow/K-theory or wonderful models who want the integral class written with explicit generators and free-cokernel arguments, and for anyone tracking agentic proof systems. The math is solid enough that a serious editor should send it to referees; the concurrent-reproduction framing and the incomplete lemma write-up are things referees can handle. I would engage with the construction and the experiment documentation.","headline":"Solid integral tangent class for matroids that reproduces Cheng's concurrent result, with one local expositional gap that does not break the construction or Hilbert identity.","tokens_in":38496,"tokens_out":556,"would_cite":false,"duration_ms":7675,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14C35","14M25","05B35","14C17"],"pacs":[],"model":"grok-4.5","headline":"An integral K-class for every loopless matroid acts like the tangent bundle of a wonderful compactification.","keywords":["matroid","wonderful compactification","building set","combinatorial K-theory","tangent class","Chow ring","Chern-alpha inequalities","Hirzebruch-Riemann-Roch"],"falsifier":"Compute the integral class for a concrete non-realizable matroid of moderate rank (for example a non-Pappus matroid with a non-maximal building set), extract its Chern numbers against powers of α, and check whether any of them falls below the corresponding binomial coefficient; or verify whether the K-theoretic Todd polynomial fails to match the known Hilbert series of the Chow ring.","tokens_in":38356,"feed_emoji":"📐","tokens_out":859,"duration_ms":7679,"temperature":0.7,"pith_summary":"The paper constructs, for every loopless matroid and every top-containing Feichtner–Yuzvinsky building set, a single integral class in the combinatorial K-ring that behaves like the tangent bundle of a De Concini–Procesi wonderful model. When the matroid is realized by a linear subspace, the class maps under a ring isomorphism to the actual tangent-bundle class of the wonderful compactification. In all cases, including non-realizable matroids, its K-theoretic Todd polynomial recovers the Hilbert series of the matroid Chow ring, and its Chern numbers against powers of the top-flat class satisfy the binomial lower bounds expected from geometry. The construction proceeds by descending a tautological quotient class from the maximal building set, first rationally via Newton power sums and then integrally by proving that one-step refinement maps have free cokernels. The result therefore supplies a uniform combinatorial substitute for a geometric tangent bundle that works whether or not a geometric model exists.","feed_headline":"Matroids get an integral tangent class that acts like geometry","feed_subtitle":"It recovers Hilbert series and Chern bounds even when no variety exists","key_machinery":"The integral tangent class T^Z_{M,G} obtained by descending the Berget–Eur–Spink–Tseng tautological quotient from the maximal building set along one-flat refinements whose maps are shown to be saturated over Z; its rationalization carries the Hilbert and Chern-alpha identities, which then lift because rationalization is a ring homomorphism fixing the boundary line classes.","core_discovery":"For every loopless matroid M of rank d+1 and every top-containing Feichtner–Yuzvinsky building set G there exists an integral class T^Z_{M,G} = sum_{F in G^circ} (1-τ_F)^{-1} - Q^Z_G in the integral combinatorial K-ring K_Z(M,G), with integer coordinates in the standard τ-monomial basis, whose rationalization is a rational tangent class that (i) specializes to the geometric tangent-bundle class of the wonderful model when M is realizable, (ii) satisfies the Hirzebruch–Riemann–Roch identity equating its K-theoretic Todd polynomial with the Hilbert series of the matroid Chow ring, and (iii) obeys the Chern-alpha inequalities deg(c_k α^{d-k}) ≥ binom(d+1,k).","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["Integral tangent class built for every loopless matroid","Matroid K-ring hosts tangent class matching wonderful models","Tangent class recovers Hilbert series of matroid Chow rings","Chern-alpha bounds hold for combinatorial tangent classes","Feichtner–Yuzvinsky sets yield integral matroid tangent classes"],"cache_read_input_tokens":32896,"weakest_assumption_plain":"The inductive step that transfers Chern-alpha numbers from a matroid to its rank-(k+1) truncation is asserted without a complete written argument in the paper; if that transfer failed, the lower bounds for ranks five and higher would lose their inductive step.","fun_headline_variants_meta":{"raw":{"variants":["Integral tangent class built for every loopless matroid","Matroid K-ring hosts tangent class matching wonderful models","Tangent class recovers Hilbert series of matroid Chow rings","Chern-alpha bounds hold for combinatorial tangent classes","Feichtner–Yuzvinsky sets yield integral matroid tangent classes"]},"model":"grok-4.5","effort":"low","cost_usd":0.004134,"raw_usage":{"total_tokens":1315,"prompt_tokens":843,"num_sources_used":0,"completion_tokens":85,"cost_in_usd_ticks":41340000,"prompt_tokens_details":{"text_tokens":843,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":387,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":843,"tokens_out":85,"duration_ms":4390,"temperature":1.0,"reasoning_tokens":387,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-11T02:03:09.783312+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Compute the integral class for a concrete non-realizable matroid of moderate rank (for example a non-Pappus matroid with a non-maximal building set), extract its Chern numbers against powers of α, and check whether any of them falls below the corresponding binomial coefficient; or verify whether the K-theoretic Todd polynomial fails to match the known Hilbert series of the Chow ring.","supporting_citations":[],"review_version":2}