{"id":"d770707e-cff0-4727-8a92-5a2b718c1970","arxiv_id":"1908.06679","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For every admissible order except 6, 7, 9 and 24, the set J3F(r) of possible 3-way flower intersection numbers equals the theoretical interval I3F(r).","lead":"The paper determines which counts of shared blocks are possible for three Steiner triple systems that all share the same 'flower' of blocks through one fixed point. It gives the complete list for nearly every admissible order, resolving a natural successor to the two-way and three-way intersection problems.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Main theorem's recursive coverage of I3F(n) is asserted rather than demonstrated; with only partial small-case sets for r=6,7,9, the equality J3F(n)=I3F(n) is not verifiable from the proof as written.","rationale":"The reader's primary worry was Lemma 4. On inspection, Lemma 4 can be proved by a standard degree-counting argument: if three systems have k+r common triples and no common flower, then for every point x the sets of noncommon triples through x in any two systems form a trade, so each point has degree at least 2 in each system's noncommon triples; this forces at least 2(2r+1)/3 noncommon triples, while the hypothesis k >= 2r(r-3)/3 allows only 4r/3. Hence the one-sentence proof is terse but the lemma is sound. The genuinely load-bearing weakness is the main theorem's interval-covering step: the proof lists five residue classes and asserts that the recursive constructions cover all values, but it never supplies the arithmetic. This is especially nontrivial because the base sets for r=6,7,9 are only partial, so even simple cases like n=19 or n=42 require checking gap-free coverage of long intervals. The reader's CONDITIONAL verdict is appropriate: the result is plausible and likely correct, but the manuscript needs an explicit covering verification or a reproducible computation before the main theorem can be accepted as proved. I therefore do not change the verdict, but I identify the interval-covering step rather than Lemma 4 as the decisive gap.","tokens_in":16649,"tokens_out":51566,"duration_ms":502194,"concrete_test":"Implement the proof's recursive constructions as a set-valued program: for each admissible n up to, say, 300, compute the set S(n) generated by Theorems 2-8 using exactly the proved small-case subsets for r=6,7,9, add the Corollary 1 high interval, and check S(n)=I3F(n). Report the first n where some element of I3F(n) is not generated; if no mismatch occurs, the asserted coverage is credible.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 20 rests on five recursive cases, but the proof does not show that the sums produced by Theorems 2-8 actually cover every integer in I3F(n). This matters because the input sets for r=6,7,9 are only partially known: for example, J3F(6) is only proved to contain [0,5]∪{7,20}, J3F(7) contains [0,8]∪[10,13]∪{16,22,28}, and J3F(9) omits several values. Case 5 for n=42 uses a B({6,7},1,42) and therefore needs the sumset 49·J3F(6)+6·J3F(7) to fill all values below the Corollary 1 threshold; that sumset has nontrivial gaps (e.g. 6 and 8 in J3F(6) are unproved), so coverage is not immediate. The paper says only that 'all required objects ... is guaranteed' and gives no interval-covering arithmetic or code. If any residue class has a hidden hole, the central equality fails. Lemma 4 itself is likely repairable: if no point had a common flower, each point would lie in at least two noncommon triples in every system, forcing at least 2(2r+1)/3 noncommon triples, contradicting the bound at most 4r/3. Thus the load-bearing gap is the unverified interval coverage in the main theorem, not Lemma 4.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the 3-way flower intersection problem for Steiner triple systems. For admissible r (r ≡ 0, 1 mod 3), J3F(r) is the set of integers k for which there exist three STS(2r+1) whose total common triples consist of k+r triples, r of them forming a common flower. The main theorem (Theorem 20) claims that J3F(n) = I3F(n) for every admissible n ≥ 10 with n ≠ 24, with partial information for n = 6, 7, 9, 24. The proof combines four recursive constructions (Latin-square constructions, a doubling construction, and PBD/GDD-based constructions), a lemma asserting that sufficiently large 3-way intersections force a common flower, and a collection of explicitly listed small base cases, some of which are checked by computer.","tokens_in":17003,"tokens_out":31361,"duration_ms":271995,"significance":"If the result is correct, it gives a complete solution to a natural 3-way analogue of the Hoffman–Lindner flower-intersection problem and extends the Milici–Quattrocchi 3-way intersection theorem to flowers. The paper has a clear recursive framework and provides a large amount of explicit small-case data, which is a genuine strength. However, two load-bearing parts of the proof are not actually demonstrated in the manuscript: Lemma 4 is justified by a one-sentence handwave, and the main theorem's recursive cases assert rather than verify the necessary set-arithmetic coverage. These gaps make the central equality J3F(n) = I3F(n) unverifiable from the proof as written, so the manuscript needs a substantial revision before it can be accepted.","major_comments":[{"comment":"Lemma 4 is load-bearing: Corollary 1, and through it nearly all high-end values in the small cases and in the main theorem, are derived from this lemma. The proof, however, consists only of the sentence 'A simple calculation now shows that there must be at least one point x for which the triples through x are the same in all systems.' No calculation is given, and the argument does not explicitly account for triples common to two of the three systems but not to the third. The lemma is very likely correct, and a counting proof is available: the condition k ≥ 2r(r−3)/3 makes the total number of non-common triple occurrences across the three systems at most 3·(4r/3) = 4r, whereas if no point has a common flower, each of the 2r+1 points is contained in at least one non-common triple in each of the three systems, giving at least 3(2r+1) = 6r+3 occurrences. This argument should be written out in full, or the lemma should be replaced by a reference to a complete proof.","section":"§2, Lemma 4"},{"comment":"The proof of the main theorem asserts, in each of the five recursive cases, that 'all required objects ... is guaranteed', but it never demonstrates that the sumsets produced by Theorems 2–8 actually cover every element of I3F(n) below the threshold of Corollary 1. This is a nontrivial arithmetic claim, especially because the input sets for r = 6, 7, 9 are only partially known. For example, in Case 5 with n = 42 (t = 7), Theorem 8 must combine 49 copies of the partial set J3F(6) ⊇ [0,5] ∪ {7,20} and 6 copies of the partial set J3F(7) ⊇ [0,8] ∪ [10,13] ∪ {16,22,28} to cover all values below 1092; the paper gives no interval-covering lemma, no explicit sumset computation, and no code for this check. Similarly, Case 1 with n = 19 and Case 2 with n = 18, 21, 27 rely on partial input sets. These checks are load-bearing: if some integer in the required interval is not covered, the equality J3F(n) = I3F(n) would fail for that n. I am not claiming the theorem is false; spot checks suggest the coverage is likely verifiable, but the proof as written does not contain the required verification.","section":"§4, proof of Theorem 20"}],"minor_comments":[{"comment":"The phrase 'It is checked by computer programming' should be supplemented with code or a machine-readable data file, or at least with a precise description of the verification procedure. The explicit systems and permutations are useful, but they are not sufficient for independent reproducibility without knowing the script used.","section":"§3, Theorems 12–14"},{"comment":"The proof of Theorem 1 says the construction is 'exactly the same' as Lemma 2.2 of Adams et al. Please provide enough detail to show how the secondary diagonal is preserved and how the parameter b is counted, since Theorem 2 depends on the exact value of b.","section":"§2, Theorem 1"},{"comment":"The definition of 'special' Latin square and the description of the cells 'above the 2×2 diagonal blocks' is informal; precise row and column indexing would make the counts in K and M easier to check.","section":"§2, Lemmas 5–7"},{"comment":"There are several small presentation issues: 'All required objects ... is guaranteed' should be 'are guaranteed', and 'F orr' at the start of Theorem 4 is a typo. More importantly, the proof should state explicitly how Theorem 1 supplies the Latin-square parameter b in Cases 1 and 3.","section":"§4, proof of Theorem 20"}],"recommendation":"major_revision","confidential_remarks":"I see no circularity or novelty problem: the external theorems are used appropriately, and the proposed result is a natural completion of previous work. The two gaps I identify are repairable, but they are central enough that the revision should be checked carefully, especially the interval-coverage arithmetic for the partial input cases."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper mostly settles the 3-way flower intersection problem for Steiner triple systems: for almost all admissible orders it pins down J3F(r), and that set was previously unknown. The result is new and, as far as I can tell, true. The machinery is standard—Latin square constructions, PBD/GDD blow-ups, small-case checking—but the assembly is genuine work, and the small cases are listed in enough detail to be checked. The citations to Milici-Quattrocchi, Hoffman-Lindner, Adams et al., and the authors' earlier KTS paper all look appropriate; I see no circularity.\n\nTwo soft spots keep me from calling the proof complete. The first is Lemma 4, where the averaging argument is compressed into one sentence. The stress-test note's sketch—if no point had a common flower, each point would lie in at least two noncommon triples in every system, forcing too many noncommon triples—is credible, so I doubt the lemma is false. But the proof as printed is not a proof. The second, more serious gap is the Main Theorem's recursive coverage step. The five cases say that the required objects are guaranteed and then stop; they never show that the sumsets produced by Theorems 2-8 actually cover every integer in I3F(n). This matters because the input sets for r=6,7,9 are only partially known. For n=42, for example, case 5 uses a B({6,7},1,42) and needs 49·J3F(6)+6·J3F(7) to fill all values below the Corollary 1 threshold. Since J3F(6) is not fully determined, the interval coverage is not immediate. It is probably still true, but the burden is on the authors to show the arithmetic or provide code. As written, the central equality is conditional on an unstated finite check.\n\nA minor note: the small-case computer checks are attributed to helpers but no code or output is included. That is common in this area and not disqualifying, but it does mean those values are also taken on faith.\n\nWho gets value from this paper: anyone working on STS intersection problems or flower intersections generally. It will be the reference for J3F(r), once the gaps are patched. It deserves a serious referee, not a desk rejection. I would send it out with a request for two things: a real proof of Lemma 4 (even a paragraph), and an explicit verification—table or code—that the recursive constructions cover the required intervals. If those land, the paper is solid.","headline":"New and mostly true result in a niche area, but the written proof leaves two load-bearing steps as assertions.","tokens_in":17488,"tokens_out":2876,"would_cite":true,"duration_ms":31742,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05B07","05B05","05B15"],"pacs":[],"model":"deepseek-v4-flash","headline":"Flower-sharing Steiner triple systems attain every allowed overlap","keywords":["Steiner triple system","3-way intersection","3-way flower intersection","Pairwise balanced design","Group divisible design","Latin square","intersection problem","recursive construction"],"falsifier":"Test Lemma 4 by computer search for small admissible orders: for $r=6$ or $r=7$, look for three STS(13) or STS(15) whose common intersection has size at least $r+2r(r-3)/3$ but in which no point has an identical flower in all three systems. One such triple of systems would break the transfer step that Corollary 1 uses for every $r\\ge 9$; alternatively, proving that some $k\\in I^3_F(n)$ with admissible $n\\ge 10$, $n\\ne 24$ cannot be realized would directly contradict Theorem 20.","tokens_in":16460,"feed_emoji":"🌸","tokens_out":13106,"duration_ms":113978,"temperature":0.7,"pith_summary":"In a Steiner triple system on $2r+1$ points, the flower at a point is the set of $r$ triples through that point. This paper determines which numbers $k$ can occur as the number of additional common triples of three Steiner triple systems that already share one whole flower, for almost every admissible order $r$ (those with $r\\equiv 0,1 \\pmod 3$). For every $r\\ge 10$ with $r\\ne 24$, the achievable set $J^3_F(r)$ equals the counting bound $I^3_F(r)=S_3[2r(r-1)/3]$, meaning all values from $0$ up to the maximum except the six values immediately below it. The orders $r=3$ and $r=4$ are settled exactly, while $r=6,7,9,24$ are left with explicit gaps between lower and upper bounds. The result matters because it shows that no subtle obstruction beyond counting prevents three systems from sharing a prescribed number of blocks together with a full flower.","feed_headline":"Three flower-sharing Steiner triple systems hit every allowed overlap","feed_subtitle":"The 3-way flower intersection problem for Steiner triple systems is solved for almost every admissible order.","key_machinery":"The central mechanism is a family of recursive constructions that build $STS(2n+1)$ out of smaller Steiner triple systems plus Latin squares. Three systems are stacked on $V\\times\\{1,2,3\\}$ and joined by Latin-square blocks (Constructions 1–3), or a system is doubled through a complete-graph $1$-factorization (Construction 4); independent control of the three inputs comes from the known 3-way intersection sets for Latin squares and for Steiner triple systems. A summation theorem (Theorem 8) then glues blocks of a pairwise balanced or group divisible design: if each block size $|B|$ admits a flower-intersection value $k_B$, the whole order admits $\\sum_B k_B$. A transfer lemma (Lemma 4, with Corollary 1) turns large ordinary 3-way intersection numbers into flower intersection numbers by asserting that sufficiently large common intersections force a point whose flower is identical in all three systems. The named object $S_3[m]$ is the target set whose six top-end gaps recur throughout the recursion.","core_discovery":"The main theorem states that for $n\\equiv 0,1 \\pmod 3$, $n\\ge 10$, $n\\ne 24$, the set $J^3_F(n)$ of 3-way flower intersection numbers coincides with $I^3_F(n)=S_3[2n(n-1)/3]$, where $S_3[m]$ is all nonnegative integers $\\le m$ except $m-1,m-2,m-3,m-4,m-5,m-7$. It also gives the complete small cases $J^3_F(3)=\\{4\\}$ and $J^3_F(4)=\\{0,8\\}$, and for $r=6,7,9,24$ it proves containment bounds that leave only short lists of values undecided. The proof constructs, for every claimed $k$, three explicit Steiner triple systems with exactly one common flower and exactly $k$ further common triples. The equality says that the full three-way analogue of the classical flower intersection problem has the same shape as the ordinary three-way intersection problem: a full interval with six top-end exceptions.","pith_inferences":["The recurring six-gap shape $S_3[m]$ across solved orders suggests that for a $\\mu$-way flower intersection problem the target set may be $S_\\mu[m]$, with the same style of top-end exceptions; the recursion here gives a template for proving such a general statement.","The unresolved cases are narrow enough that a finite exhaustive search could close them: only one value ($k=6$) is missing for $r=6$, and only $\\{9,14,15\\}$ for $r=7$, given the paper's bounds.","The same Latin-square stacking and PBD summation machinery should transfer to flower intersection problems for other designs, such as $S(2,4,v)$ or Kirkman triple systems with a prescribed parallel class, where analogous ingredients are known."],"forward_implications":["All admissible orders $n\\ge 10$ except $n=24$ have $J^3_F(n)=I^3_F(n)$, so every allowed value from $0$ to $2n(n-1)/3$, apart from the six top-end exceptions, is realized by three flower-sharing systems.","The undecided orders are exactly $n=6,7,9,24$, with only short explicit value lists left open: for instance $k=6$ for $n=6$ and $k\\in\\{9,14,15\\}$ for $n=7$.","The induction rests on small bases $n=3,4,10,15,60,132$; once these are available in the right residue classes, the PBD/GDD gluing produces every larger case.","The maximum value $2n(n-1)/3$ is always realized by taking three identical systems, and $0$ is realized for all $n\\ge 4$, so the extreme ends of the spectrum are never exceptional."],"supporting_citations":[{"why":"Supplies the complete solution of the 3-way intersection problem for Latin squares, the input used in Constructions 1–3 and Theorems 2–4.","marker":"Adams et al. (2002)"},{"why":"Determines the sets $J^3(v)$ for three Steiner triple systems, giving upper bounds and small base systems for the recursion.","marker":"Milici and Quattrocchi (1987)"},{"why":"Solves the two-system flower intersection problem $J_F(r)$ and supplies the lemma pattern from which Lemma 4 is adapted.","marker":"Hoffman and Lindner (1987)"},{"why":"Provides the classical two-system intersection set $J(v)$ that the flower and three-way problems generalize.","marker":"Lindner and Rosa (1975)"},{"why":"Gives large sets of disjoint 3-GDDs of type $2^r4^1$, used to prove $0\\in J^3_F(r)$ for every admissible $r\\ge 4$.","marker":"Cao et al. (2001)"},{"why":"Supplies the $\\{4\\}$-GDD of type $3^4 6^2$ used to settle the order $r=24$ through the summation theorem.","marker":"Kreher and Stinson (1997)"},{"why":"Supplies the $\\{4\\}$-GDD of type $6^{10}$ used for the order $r=60$.","marker":"Wei and Ge (2014)"},{"why":"Supplies the $\\{4\\}$-GDD of type $9^{12}24^1$ used for the order $r=132$.","marker":"Wei and Ge (2013)"}],"fun_headline_variants":["3-flower overlaps in STS: full range minus six","STS: 3-way flower intersection almost solved","Three STS sharing a flower: all counts but six top","Flower overlaps in STS: resolved for all but four orders"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is Lemma 4: whenever three Steiner triple systems of order $2r+1$ share at least $r+2r(r-3)/3$ triples, some point has the same flower in all three systems. The paper supports this with a one-sentence 'simple calculation' sketch, and Corollary 1 plus every large-order case built from it depends on that calculation being correct.","fun_headline_variants_meta":{"raw":{"variants":["3-flower overlaps in STS: full range minus six","STS: 3-way flower intersection almost solved","Three STS sharing a flower: all counts but six top","Flower overlaps in STS: resolved for all but four orders"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000583,"raw_usage":{"total_tokens":2742,"prompt_tokens":945,"completion_tokens":1797,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":561,"completion_tokens_details":{"reasoning_tokens":1731}},"tokens_in":561,"tokens_out":1797,"duration_ms":16431,"temperature":1.0,"reasoning_tokens":1731,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:38:37.968132+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Test Lemma 4 by computer search for small admissible orders: for $r=6$ or $r=7$, look for three STS(13) or STS(15) whose common intersection has size at least $r+2r(r-3)/3$ but in which no point has an identical flower in all three systems. One such triple of systems would break the transfer step that Corollary 1 uses for every $r\\ge 9$; alternatively, proving that some $k\\in I^3_F(n)$ with admissible $n\\ge 10$, $n\\ne 24$ cannot be realized would directly contradict Theorem 20.","supporting_citations":[],"review_version":1}