{"id":"bc1ed453-5bd1-4b25-bb18-48563a4abe98","arxiv_id":"2607.08446","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":3,"one_line_summary":"Monte Carlo methods (pair matching, reference subsets, weight-layer branching) give D(10)≈8.93×10^78 through D(15)≈3.81×10^1953 with explicit standard errors, beating Korshunov asymptotics.","lead":"Researchers statistically estimate Dedekind numbers D(10) through D(15) to a few significant digits using Monte Carlo sampling of monotone Boolean functions. The numbers grow so fast that exact values beyond D(9) remain out of reach, so better estimates matter for testing algorithms and asymptotics in combinatorics.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"For n=15 the reported midpoint discrepancy (2.9%) exceeds the statistical SE (~0.14%) by ~20\times, so the claimed 2-digit precision and SE for D(13)–D(15) rest on an unquantified systematic component of the weight-layer estimator.","rationale":"The reader correctly isolates the mixing/branching assumption for n≥13 as the weakest link and already assigns CONDITIONAL. The concrete numerical mismatch between Table-14 discrepancy and SE simply makes that assumption sharper: the internal consistency diagnostic already signals that the quoted error bars understate total uncertainty for the largest n. No other load-bearing flaw (circularity, missing code, contradiction with the concurrent Chen et al. D(10) result, or failure of the three-method agreement on D(10)–D(12)) appears. The verdict therefore stays CONDITIONAL; the only adjustment is a more precise statement of why the higher-n figures remain provisional.","tokens_in":15907,"tokens_out":599,"duration_ms":22799,"concrete_test":"Extract the two independent middle-layer cardinality estimates S^{(0)}_{2^{n-1}} and S^{(2^n)}_{2^{n-1}} for the n=15 runs already performed (or re-run with the same total core-days). If their relative difference stays ~0.029 while the multi-run SE remains ~0.0014, replace the published SE by max(statistical SE, discrepancy) and re-state the number of reliable digits; a material drop below two digits would weaken the D(15) headline.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central numerical claims for D(13)–D(15) rest solely on weight-layer branching (Section 4.4). Upward and downward propagations of degree-corrected left/right branching ratios meet at the middle layer; their relative discrepancy is reported in Table 14 and grows from ~10^{-4} (n=10) to 2.90×10^{-2} (n=15). The SE, however, is obtained only from run-to-run variation of the ratios and is taken as the relative error on the most populated layer. For n=15 this yields relative SE ≈1.4×10^{-3}, an order of magnitude smaller than the observed discrepancy. The paper therefore presents a consistency check that is larger than the quoted uncertainty, indicating that residual bias (incomplete local mixing on the middle layers, or imperfect degree correction inside the two-layer subgraphs) is not folded into the reported SE. The multi-method cross-checks that underwrite D(10)–D(12) are unavailable here, so the 2-digit claim for D(15) is the least secure part of the strongest claim.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The manuscript reports Monte Carlo estimates of the Dedekind numbers D(10)–D(15), improving substantially on Korshunov’s asymptotics and on concurrent layer-ratio work for D(10). Three estimators are used: (i) pair-matching frequency of uniform MBF9 samples to obtain D(10); (ii) MCMC hit rates into closed-form 1-layer reference subsets for D(10)–D(12); and (iii) degree-corrected weight-layer branching, which supplies estimates for all of D(10)–D(15). For n=9 the MCMC is validated against large uniform pair-matched samples (KS p=0.29 after 30k burn-in); Hamming-distance and weight-balance diagnostics are given for n=9–10. Multi-method agreement within reported S.E. is shown for D(10)–D(12) (Tables 11–13). For D(13)–D(15) only weight-layer branching is practical; Table 14 lists values, run-to-run S.E., core-day runtimes, and midpoint discrepancies between upward and downward layer propagations.","tokens_in":16241,"tokens_out":1526,"duration_ms":19758,"significance":"Exact Dedekind numbers are known only through D(9); higher values are of lasting interest in enumerative combinatorics and order theory. The paper supplies the first estimates with explicit numerical uncertainties for D(10)–D(15), multi-method cross-checks where feasible, open implementations, and a clear improvement over the only previously available asymptotic formulae (errors of tens of percent). The concurrent independent D(10) estimate of Chen et al. is consistent with the present results, which strengthens confidence in the n=10 figure. If the higher-n error budgets are made rigorous, the tables will become standard reference values for the community.","major_comments":[{"comment":"Section 4.4 and Table 14: the reported S.E. for D(13)–D(15) is taken solely from run-to-run variation of the branching ratios (relative error on the most populated layer). For n=15 the midpoint discrepancy between the independent upward and downward propagations is 2.90×10^{-2}, roughly twenty times larger than the quoted relative S.E. (~1.4×10^{-3}). The discrepancy is therefore a consistency diagnostic that exceeds the published uncertainty, indicating residual systematic bias (incomplete local mixing on middle layers, or imperfect degree correction inside the two-layer subgraphs) that is not folded into the error budget. The abstract’s claim of “2 digits for D(15)” and the S.E. column of Table 1 rest on this unquantified component. Either enlarge the uncertainty to cover the observed discrepancy (or a calibrated multiple of it), or supply independent diagnostics that demonstrate the d","section":"Section 4.4, Table 14"},{"comment":"Sections 3.4.2–3.5 validate mixing (Hamming distance, weight-balance, KS against uniform samples) only for n=9 and, partially, n=10. For n≥13 the estimator relies entirely on degree-corrected walks on adjacent weight layers (Section 4.4). The paper should state what burn-in, thinning, and local-mixing checks were used at those dimensions, or acknowledge that the n=9–10 diagnostics are being extrapolated. Without this, the assumption that the left/right branching ratios faithfully represent true layer cardinalities remains the weakest link for the D(13)–D(15) claims.","section":"Sections 3.4.2–3.5, 4.4"},{"comment":"Table 1 and the abstract present a single “best” value and S.E. for each n. For D(10)–D(12) the three methods differ by amounts comparable to (or larger than) the smallest quoted S.E. (e.g., reference-subset D(12)=7.1911×10^{283} vs weight-layer 7.1492×10^{283}). The final reported figure should either be a variance-weighted combination with an enlarged uncertainty that reflects method-to-method scatter, or the text should explicitly justify why one method’s S.E. is preferred over the inter-method spread.","section":"Table 1, Tables 11–13"}],"minor_comments":[{"comment":"Table 3 caption states that relative errors for D(10)–D(15) are “relative to this paper’s estimates … and [are] therefore [themselves] uncertain.” That is correct, but the table still prints those percentages in the same column as exact-error percentages for D(0)–D(9). A visual distinction (e.g., italics or a second column) would avoid over-reading the higher-n entries.","section":"Table 3"},{"comment":"Equation (13)–(14) and Table 9: p_match,n for n≥9 are derived backwards from the paper’s own D(n) estimates. The table header “Best currently known values” is slightly misleading for those rows; label them as “inferred from Table 1” to avoid circular appearance.","section":"Table 9, Theorem 3"},{"comment":"Figure 2 caption: “The relative height between series has no meaning” is clear, but the arbitrary vertical scaling constants are not stated; a short note that each curve is independently normalised would help reproducibility of the figure.","section":"Figure 2"},{"comment":"Appendix A.15 reports average antichain sizes up to n=15 with S.E.; the n=15 entry has S.E. 0.767 on a mean of ~2295, which is fine, but the text never uses these numbers outside the filter-tree discussion. Either cite them in the main MCMC section or move the higher-n rows to a repository note.","section":"Table A.15"},{"comment":"Minor typos: “soft page faults” (p. 5) is fine; “unneeded MBF6 samplings” → “unnecessary”; “the bulk of the time” is colloquial for a journal. “S.E.2.44\times10^{74}” in Table 1 needs a space after “S.E.”.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The concurrent arXiv:2606.09795 (Chen et al.) is properly cited and the D(10) numbers agree; no priority dispute is apparent. The work is a good fit for a combinatorics or computational-mathematics venue. The main risk is overstated precision for D(13)–D(15); once the error budget is honest the paper should be publishable. Code links are given and appear central to reproducibility."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The real news is usable multi-digit numbers for D(10)–D(15) with standard errors, obtained by three independent Monte Carlo pipelines that agree where they can be compared. That is a genuine step past Korshunov’s asymptotics and past the single concurrent D(10) estimate of Chen et al.\n\nWhat is new is the weight-layer branching estimator (propagating degree-corrected left/right branching ratios from the two ends of the weight graph until they meet at the middle layer) plus large-scale pair-matching on a 2.3 B MBF9 buffer and reference-subset sampling against the closed-form 1-layer middle-layer sets. For D(10) the three methods sit inside each other’s SEs (pair-matching ~8.938×10^78, weight-layer ~8.937×10^78, reference-subset 8.933×10^78 with the tightest SE 2.44×10^74). D(11) and D(12) still have two-method agreement. MCMC is validated against uniform pair-matched MBF9s (KS p=0.29 after 30 k burn-in), Hamming-distance and weight-balance diagnostics are given for n=9–10, code is public, and the known exact values through D(9) are used only as inputs and checks. That is clean computational combinatorics.\n\nThe soft spot is exactly the one the stress-test flags, and it is real but limited. For n≥13 only weight-layer branching remains practical. Table 14 shows the midpoint discrepancy growing from ~10^{-4} at n=10 to 2.9 % at n=15, while the reported relative SE (run-to-run variation of the ratios, taken on the densest layer) is ~0.14 %. The paper therefore quotes a statistical error smaller than its own consistency check; residual bias from incomplete local mixing or imperfect degree correction inside the two-layer subgraphs is not folded into the SE. The authors disclose the discrepancy and do not hide the single-method status, so this is not opacity—just an honest limit on how many digits one should trust for D(13)–D(15).\n\nThe paper is for people who need concrete targets for Dedekind numbers, antichain enumeration, or MCMC on posets. The math and citation pattern look solid; free parameters (burn-in, filter-tree threshold, sample sizes) are ordinary Monte Carlo knobs, not free fits. I would send it to referees. They will ask for a more conservative error budget on the higher-n numbers, but the core advance is real and reproducible.","headline":"Solid multi-method Monte Carlo estimates for D(10)–D(15) that beat Korshunov and match concurrent D(10) work; the soft spot is that for n≥13 the midpoint discrepancy exceeds the quoted SE, so the 2-digit claims rest on an unquantified systematic.","tokens_in":16872,"tokens_out":666,"would_cite":true,"duration_ms":6509,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["06A07","05A15","68R05","68W20","65C05"],"pacs":[],"model":"grok-4.5","headline":"Statistical Monte Carlo methods give D(10)–D(15) to 4–2 significant digits, far tighter than prior asymptotics.","keywords":["Dedekind numbers","monotone Boolean functions","antichains","Monte Carlo methods","MCMC","combinatorial enumeration","weight layer branching"],"falsifier":"An independent high-precision computation of D(10)—exact enumeration or a wholly different Monte Carlo scheme—that lands outside the interval 8.93345×10^78 ± a few reported standard errors would falsify the central numerical claims.","tokens_in":16775,"feed_emoji":"🔢","tokens_out":899,"duration_ms":22922,"temperature":0.7,"pith_summary":"Dedekind numbers D(n) count the monotone Boolean functions on n variables; exact values stop at D(9) because the sequence grows double-exponentially. This paper supplies the first high-accuracy statistical estimates for D(10) through D(15), reporting four reliable digits for D(10) and two for D(15). Three independent sampling techniques—pair matching of random 9-variable functions, reference-subset hit rates, and weight-layer branching—are used; where more than one method applies they agree inside their standard errors. The results replace Korshunov’s asymptotic formulas, which err by tens of percent even on known values, and give concrete numerical targets for future exact or improved Monte Carlo work. A sympathetic reader cares because Dedekind’s problem is a classical open enumeration challenge whose next few terms have been out of reach for decades.","feed_headline":"Dedekind numbers D(10)–D(15) estimated to 2–4 digits","feed_subtitle":"Monte Carlo walks on monotone Boolean functions beat old asymptotics by orders of magnitude","key_machinery":"Weight Layer Branching: degree-corrected random walks confined to adjacent weight layers of the MBF graph measure average left/right branching ratios; the ratios are propagated from the constant-0 and constant-1 functions to the middle layer, producing estimates of every layer cardinality and therefore of D(n).","core_discovery":"Weight-layer branching, cross-checked by pair matching and reference-subset sampling wherever feasible, yields D(10)≈8.93345×10^78 (S.E. 2.44×10^74) through D(15)≈3.80603×10^1953 (S.E. 5.30×10^1951). Multi-method consistency for n=10–12 and midpoint discrepancies of order 10^{-4}–10^{-2} for higher n support the claim that these figures are accurate to the stated precision and substantially better than earlier estimates.","pith_inferences":["Middle-layer concentration of min-positive and max-negative sets for n>9 suggests asymptotic formulas can treat only those layers as free variables.","Degree-corrected MCMC on the MBF graph is likely transferable to free distributive lattices and other graded posets of similar growth.","If midpoint discrepancy grows only slowly, the method can reach D(16)–D(18) with feasible core-years before exact enumeration becomes realistic.","The sequence of pair-matching probabilities p_match,n itself may admit an independent asymptotic analysis."],"forward_implications":["Four-digit D(10) and two-digit D(15) replace Korshunov formulas whose relative errors exceeded 50 percent on known values.","Agreement among three methods for D(10)–D(12) cross-validates the MCMC mixing diagnostics and degree correction.","Midpoint discrepancies of order 10^{-4}–10^{-2} supply a concrete residual-error diagnostic for the higher estimates.","The same sampling infrastructure immediately supports the larger reference subsets and Metropolis–Hastings variants listed as future work.","Layer probabilities obtained en route yield an independent formula for expected Hamming distance between random MBFs."],"fun_headline_variants":["Weight-layer branching estimates Dedekind D(10)–D(15) to 2–4 digits","Monotone Boolean sampling refines Dedekind numbers D(10) through D(15)","Multi-method Monte Carlo yields D(10)≈8.9e78 to D(15)≈3.8e1953","Pair matching and reference subsets cross-check D(10)–D(12) estimates","Higher Dedekind numbers D(10)–D(15) estimated beyond prior asymptotics"],"cache_read_input_tokens":128,"weakest_assumption_plain":"For n≥13 the estimates rest on the premise that short degree-corrected walks between neighbouring weight layers have mixed enough for the observed branching ratios to represent the true global layer sizes.","fun_headline_variants_meta":{"raw":{"variants":["Weight-layer branching estimates Dedekind D(10)–D(15) to 2–4 digits","Monotone Boolean sampling refines Dedekind numbers D(10) through D(15)","Multi-method Monte Carlo yields D(10)≈8.9e78 to D(15)≈3.8e1953","Pair matching and reference subsets cross-check D(10)–D(12) estimates","Higher Dedekind numbers D(10)–D(15) estimated beyond prior asymptotics"]},"model":"grok-4.5","effort":"low","cost_usd":0.008922,"raw_usage":{"total_tokens":1997,"prompt_tokens":717,"num_sources_used":0,"completion_tokens":131,"cost_in_usd_ticks":89220000,"prompt_tokens_details":{"text_tokens":717,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1149,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":717,"tokens_out":131,"duration_ms":10172,"temperature":1.0,"reasoning_tokens":1149,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-10T07:20:04.878825+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"An independent high-precision computation of D(10)—exact enumeration or a wholly different Monte Carlo scheme—that lands outside the interval 8.93345×10^78 ± a few reported standard errors would falsify the central numerical claims.","supporting_citations":[],"review_version":1}