{"id":"c66f7752-e5c9-4215-af40-c6d60a28f704","arxiv_id":"2507.20932","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Continuous distributors of sites are shown to be equivalent to geometric morphisms between the associated sheaf topoi, unifying morphisms and comorphisms of sites.","lead":"This paper introduces a notion of distributor of sites that generalizes both morphisms and comorphisms of sites, and proves that the continuous ones correspond exactly to geometric morphisms between the associated sheaf topoi. The result gives topos theorists a unified relational language for comparing sites and their sheaf topoi.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 2.1.13 silently assumes cover-distributivity; the main theorem's proof uses the lemma only for distributors of sites, so the claim likely survives a corrected lemma, but the proof as written is incomplete.","rationale":"The reader's weakest assumption identifies exactly the gap I would stress: Lemma 2.1.13 is stated for arbitrary distributors but its proof uses cover-distributivity, which is not part of the statement. This is the most load-bearing defect because the proof of the main equivalence Theorem 2.2.19 relies on the lemma to convert flatness of f∗aJ yC into K-flatness of Hf, and to show that every distributor of sites gives a lex inverse image. I checked the surrounding argument in good faith: the theorem is applied only to distributors of sites, which do satisfy the missing hypothesis, and the continuous-distributor condition supplies the sheaf-valuedness needed for the equivalence to be an equivalence rather than an adjunction. So the central claim is likely correct once the lemma is corrected, but the written proof is incomplete. The reader's CONDITIONAL verdict is therefore appropriate, and my stress-test does not move it.","tokens_in":20536,"tokens_out":58210,"duration_ms":619624,"concrete_test":"Restate Lemma 2.1.13 with the explicit hypothesis that H is cover-distributing from J to K, and re-derive the three-way equivalence using Lemma 2.1.10 for the implication K-flat ⇒ aK bH flat. Then audit every invocation of Lemma 2.1.13 in §2.2—especially Proposition 2.2.11 and Lemma 2.2.3—and verify that the cover-distributing hypothesis is satisfied in each case. If any invocation relies on the lemma without this hypothesis, the proof of Theorem 2.2.19 must be patched before the equivalence can be accepted as proven.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central equivalence Theorem 2.2.19 depends on Lemma 2.1.13, which states that for an arbitrary distributor H : C ↬ D, K-flatness is equivalent to flatness of aK bH : C → bDK and to lexness of aKlextH iJ : bCJ → bDK. The proof of the lemma begins 'since H is in particular cover distributive', an assumption that appears neither in the statement of Lemma 2.1.13 nor in Definition 2.1.11 of K-flatness. The missing hypothesis is not cosmetic: the argument needs Lemma 2.1.10 to know that aK bH sends J-covers to epimorphic families, and this is exactly what cover-distributivity provides. Without it, aKlextH need not factor through aJ, so the composite aKlextH iJ is not the induced inverse image functor on sheaves, and its lexness is not equivalent to the flatness of aK bH. This is the load-bearing gap because Proposition 2.2.11 uses Lemma 2.1.13 twice: to prove that the distributor Hf extracted from a geometric morphism is K-flat, and to prove that a distributor of sites induces a lex inverse image. In all such applications the distributor under consideration is cover-distributing by construction, so the main theorem is probably repairable by adding 'H cover-distributing' to the lemma. As written, however, the proof of the central claim has an unstated hypothesis.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a notion of 'distributor of sites' between two sites (C,J) and (D,K), defined as a distributor H : C ↬ D that is cover-distributing from J to K and K-flat. It shows that such distributors induce geometric morphisms between the associated sheaf topoi, and that every geometric morphism arises from such a distributor (Proposition 2.2.11). A further 'continuity' condition on distributors (bH taking values in the sheaf topos) is introduced, and the paper claims in Theorem 2.2.19 an equivalence of categories Top[ bDK, bCJ ] ≃ SitW_Cont[(C,J),(D,K)], refining the adjunction of Proposition 2.2.11. The paper also discusses a bicategory SitW of sites and distributors of sites, its 'equipment-like' properties, gluing constructions, and connections to join-approximable relations and toposes.","tokens_in":20908,"tokens_out":7659,"duration_ms":86495,"significance":"If the main equivalence (Theorem 2.2.19) is correct, this is a substantial contribution: it provides a single profunctorial notion that generalizes both morphisms and comorphisms of sites, and it gives a classification of all geometric morphisms between sheaf topoi in terms of continuous distributors. The approach is natural and builds on standard tools (Diaconescu's theorem, flatness criteria, sheafification), and the paper makes a plausible case that the construction is pseudofunctorial. The paper also offers several useful auxiliary results, such as the gluing proposition (Proposition 2.3.5). However, the manuscript as written has a gap in a key lemma (Lemma 2.1.13) and, more importantly, the central theorem is stated without a proof. The overall framework is promising, but the paper needs substantial revision before the claims can be accepted.","major_comments":[{"comment":"The lemma states that for an arbitrary distributor H : C ↬ D, K-flatness is equivalent to flatness of aK bH and to lexness of aKlextH iJ, but the proof uses the hypothesis that H is cover-distributing ('since H is in particular cover distributive'). This hypothesis is not in the statement and is not part of Definition 2.1.11 of K-flatness. Moreover, the third equivalent condition (lexness of aKlextH iJ : bCJ → bDK) is only meaningful when aKlextH factors through aJ, which is exactly what cover-distributivity ensures via Lemma 2.1.10. Without that hypothesis, the composite aKlextH iJ may not be the induced inverse image functor on sheaves, so the stated equivalence is not correct. The fix is local: add 'H cover-distributing from J to K' to the lemma (or split the lemma into the unconditionally valid equivalence between K-flatness and flatness of aK bH, plus a separate statement for the lexness condition under cover-distributivity). Since all applications in §2.2 concern distributors of sites, which are cover-distributing by definition, the main theorem is likely repairable, but the proof as written is incomplete.","section":"§2.1, Lemma 2.1.13"},{"comment":"The central result of the paper is stated without proof. The text preceding the theorem says that SitW_Cont[(C,J),(D,K)] is a reflective subcategory of SitW and that the adjunction of Proposition 2.2.11 restricts to an equivalence, but this is an assertion rather than a proof. A rigorous proof should show that for every continuous distributor H the unit H → H_{Sh(H)} is invertible (which follows if bH already lands in bDK), and that for every geometric morphism f the extracted distributor H_f is continuous and Sh(H_f) ≃ f. These steps are not written down, and because Theorem 2.2.19 is the main claim of the paper, this is a load-bearing gap.","section":"§2.2, Theorem 2.2.19"},{"comment":"The proof of the converse direction ('Conversely suppose that H is Johnstone-Wraith continuous...') is only a sketch. In particular, the statement 'applying this fact to bidense morphisms into finite limits, aKlextH is exhibited as being lex, hence H to be K-flat' is not justified in detail, and the passage from Johnstone-Wraith continuity to K-flatness requires care because the definition of K-flatness involves arbitrary pairs and parallel pairs, not just finite limits of representables. Since this proposition is not used in the proof of Theorem 2.2.19, it is not a blocker, but it should be expanded for the paper to be self-contained.","section":"§2.2, Proposition 2.2.9"}],"minor_comments":[{"comment":"The notation ContJ[C, bDK] is used in the displayed sequence of isomorphisms but is never defined; please define it (presumably the category of functors C → bDK that send J-covers to epimorphic families).","section":"§2.2, Proposition 2.2.11"},{"comment":"In Definition 2.2.12, the phrase 'bH : C → bD takes values in the sheaf topos bDK' is slightly imprecise, since the codomain of bH is bD; it would be clearer to say that bH factors through the inclusion iK : bDK → bD.","section":"§2.2, Definition 2.2.12"},{"comment":"There is a typo: 'continuous ditributors' should be 'continuous distributors'.","section":"§2.2, Example 2.2.24"},{"comment":"The sentence 'by [13][VII.9 theorem 1] the functor aK bH is filtering if and only if it is flat, and this amounts for the induced aKlextH to being lex' is grammatically unclear and would benefit from being split into two sentences.","section":"§2.1, Lemma 2.1.13, proof"}],"recommendation":"major_revision","confidential_remarks":"The paper is likely correct in its main idea, and the gap in Lemma 2.1.13 is easily fixed by adding the missing cover-distributivity hypothesis. However, the absence of a proof of Theorem 2.2.19 is more serious: the central equivalence is the paper's headline result, and it is merely asserted. I recommend major revision with the expectation that the authors supply a complete proof of the main theorem and correct the lemma. The paper will then be a valuable contribution to the theory of sites and topoi."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear Colleague,\n\nThe short version: this is a solid topos theory paper that introduces a genuinely useful notion, but the proof of Lemma 2.1.13 is incomplete as written. I think the main theorem survives, but a referee should ask for a fix.\n\nWhat's new: the definition of distributor of sites (cover-distributing + K-flat) and the adjunction/equivalence between continuous distributors and geometric morphisms. That equivalence, Theorem 2.2.19, is a real result and unifies the two kinds of maps between sites. The paper also handles the subtlety that continuous distributors don't compose, and proposes a modified bicategory SitW_Cont. The gluing/equipment discussion at the end is a nice bonus, though it's more exploratory.\n\nWhat's good: the authors are pretty honest about the relation to prior work. Proposition 2.2.9 connects their notion to Johnstone-Wraith's bidense continuity, which keeps the novelty claim calibrated. The hard direction, that every geometric morphism yields a distributor of sites, is grounded in Diaconescu's theorem rather than being engineered by definition. That's the right kind of argument.\n\nThe soft spot is Lemma 2.1.13. It states an equivalence for an arbitrary distributor H, but the proof says 'since H is in particular cover distributive' — an assumption that appears in neither the statement nor the definition of K-flatness. This isn't cosmetic: without cover-distributivity, aK bH need not send J-covers to epimorphic families, and the composite aKlextH iJ need not factor through aJ. So the lemma as stated is false or at least unproved. That said, the applications in Proposition 2.2.11 and Theorem 2.2.19 only use the lemma for distributors of sites, which are cover-distributing by definition. So the central claim is probably repairable by adding 'H cover-distributing' to the lemma. A referee should verify the proof goes through with that hypothesis.\n\nOther quibbles: the abstract and introduction say 'sites' where the theorem needs 'small generated' sites; that's an overstatement. Some proofs, like Proposition 2.2.9, are sketches rather than full arguments. But these are minor relative to the main gap.\n\nBottom line: if you work in topos theory or Caramello's bridge program, this is worth reading and citing. It deserves a serious referee — the main idea is good and the flaw is localized. I'd recommend sending it to review with a request to fix Lemma 2.1.13 and tighten the small-generated hypotheses.\n\nBest,","headline":"Useful unification of morphisms and comorphisms of sites via distributors; central theorem likely correct but Lemma 2.1.13 has a gap as stated.","tokens_in":21435,"tokens_out":2371,"would_cite":true,"duration_ms":24315,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["18B25","18F10"],"pacs":[],"model":"deepseek-v4-flash","headline":"Every geometric morphism between sheaf topoi is induced, uniquely up to isomorphism, by a continuous distributor of sites.","keywords":["distributors of sites","geometric morphisms","sheaf topoi","cover-flatness","cover-distributivity","morphisms of sites","comorphisms of sites","proarrow equipments"],"falsifier":"Test Lemma 2.1.13 on a distributor that is $K$-flat but not cover-distributing, for example one induced by a flat but not cover-preserving functor: if $a_K \\mathrm{lext}_H i_J$ is not lex, then the lemma as stated is false. To test Theorem 2.2.19, look for two continuous distributors between the same sites that induce isomorphic geometric morphisms but are not themselves isomorphic; the theorem predicts none exist, since the equivalence is categorical.","tokens_in":20352,"feed_emoji":"🔗","tokens_out":7617,"duration_ms":81265,"temperature":0.7,"pith_summary":"This paper introduces distributors of sites, a single kind of arrow between sites that contains both morphisms and comorphisms of sites as special cases. The central result is that, for small generated sites, continuous distributors of sites are in equivalence with geometric morphisms between the associated sheaf topoi: every geometric morphism is induced by a continuous distributor, uniquely up to isomorphism. The authors also show that the bicategory of sites and distributors of sites has several equipment-like features, including gluing of distributors, and that the continuous variant is the Kleisli bicategory of the sheafification pseudomonad. A sympathetic reader should care because the result gives a site-level, relational calculus for all morphisms of sheaf topoi, not only those coming from functors.","feed_headline":"All geometric morphisms arise from site distributors","feed_subtitle":"A single relational notion subsumes morphisms and comorphisms of sites and matches sheaf-topos maps one-to-one.","key_machinery":"The central object is a distributor of sites: a distributor $H : C \\looparrowright D$, that is, a functor $D^{\\mathrm{op}} \\times C \\to \\mathrm{Set}$ thought of as heteromorphisms, equipped with two site-relative conditions. Cover-distributivity says that pulling back the image $H[S]$ of any $J$-covering sieve $S$ along any heteromorphism $x \\in H(d,c)$ gives a $K$-covering sieve on $d$; $K$-flatness is the coverage-relative version of flatness, saying that $a_K \\widehat{H} : C \\to \\widehat{D}_K$ is a flat functor, equivalently that the sheafified extension $a_K \\mathrm{lext}_H i_J$ is lex. Continuity asks that $\\widehat{H}$ take values in sheaves. The argument is carried by the composite $\\mathrm{Sh}(H)^* = a_K \\mathrm{lext}_H i_J$, which sends a distributor to the inverse image of a geometric morphism; compatibility of this construction with composition is what produces the equivalence.","core_discovery":"The paper's main discovery is Theorem 2.2.19: for any small generated sites $(C,J)$ and $(D,K)$ there is an equivalence of categories $\\mathrm{Top}[\\widehat{D}_K,\\widehat{C}_J] \\simeq \\mathrm{SitW}_{\\mathrm{Cont}}[(C,J),(D,K)]$. In plain terms, the geometric morphisms between the sheaf topoi are classified by the continuous distributors of sites between the sites themselves. A distributor of sites is a distributor $H : C \\looparrowright D$ that is cover-distributing from $J$ to $K$ and $K$-flat, and it is continuous when the induced functor $\\widehat{H} : C \\to \\widehat{D}$ lands in the sheaf topos $\\widehat{D}_K$. The equivalence makes every geometric morphism $f$ arise as $\\mathrm{Sh}(H_f)$ from the continuous distributor $H_f(d,c) = f^*(a_J \\yo_c)(d)$, and it recovers the classical picture: representable distributors $D[1,f]$ give morphisms of sites, while corepresentable distributors $C[f,1]$ give comorphisms of sites.","pith_inferences":["Editorial inference: since the equivalence is phrased entirely in terms of sites and distributors, it supplies a calculus in which a geometric morphism can be manipulated as a relation of heteromorphisms; questions such as whether a morphism of topoi preserves a given logical structure could become concrete sieve conditions on the sites.","Editorial inference: the analogy drawn with join-approximable relations suggests a broader pattern, namely that distributors of sites may provide Stone-type dualities for other doctrines; this avenue is left open by the paper.","Editorial inference: because continuous distributors do not compose inside $\\mathrm{SitW}$, the paper effectively constructs a 2-localization of the bicategory of sites and distributors; testing whether this localization has all the structure of a proarrow equipment once right adjoints are allowed in a suitably enlarged category would be a natural next step."],"forward_implications":["Morphisms of sites and comorphisms of sites are both recovered as distributors of sites: $D[1,f]$ for a morphism $f$ and $C[f,1]$ for a comorphism $f$ (Theorem 2.2.23).","Every geometric morphism between sheaf topoi admits a site-level presentation as a continuous distributor, even when no functorial lift exists between the fixed sites (Corollary 2.2.20).","Continuous distributors of sites form a bicategory $\\mathrm{SitW}_{\\mathrm{Cont}}$ with modified units and composition, and $\\mathrm{Sh}$ factors through it; this bicategory is the Kleisli bicategory of the sheafification pseudomonad (Proposition 2.2.18 and 2.3.1).","A distributor of sites induces a geometric surjection exactly when it is cover-testing, a heteromorphic analogue of cover-reflecting (Proposition 2.2.27).","Distributors of sites can be glued into a site whose two inclusions are simultaneously morphisms and comorphisms of sites, giving cotabulators universal on both sides (Proposition 2.3.5)."],"supporting_citations":[{"why":"Supplies the notion and basic theory of flat distributors, the foundation on which the paper's site-relative flatness is built.","marker":"[2]"},{"why":"Provides the flat-functor correspondence and filteredness criteria used to link flat distributors to lex inverse images and to extract distributors from geometric morphisms.","marker":"[13]"},{"why":"Contributes the bidense-morphism continuity condition used to characterize distributors of sites in Proposition 2.2.9.","marker":"[10]"},{"why":"The companion first part of this program, whose double-categorical framework for morphisms and comorphisms the present distributor notion is designed to subsume.","marker":"[8]"},{"why":"Gives the density criterion for geometric surjections that the paper adapts to distributors in Proposition 2.2.27.","marker":"[7]"},{"why":"Supplies the definition of proarrow equipment used to discuss the equipment-like properties of the bicategory of sites and distributors.","marker":"[16]"}],"fun_headline_variants":["Site distributors unify morphisms and comorphisms","Geometric morphisms arise from continuous site distributors","One notion of site distributor captures all geometric maps","Distributors of sites classify sheaf-topos morphisms","Site distributors: a single concept for site maps and topoi"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that $K$-flatness of a distributor is equivalent to lexness of its sheafified extension $a_K \\mathrm{lext}_H i_J$; the proof of Lemma 2.1.13 silently assumes the distributor is cover-distributing, a hypothesis absent from the lemma's statement but satisfied by the distributors of sites to which the main theorem applies.","fun_headline_variants_meta":{"raw":{"variants":["Site distributors unify morphisms and comorphisms","Geometric morphisms arise from continuous site distributors","One notion of site distributor captures all geometric maps","Distributors of sites classify sheaf-topos morphisms","Site distributors: a single concept for site maps and topoi"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000587,"raw_usage":{"total_tokens":2721,"prompt_tokens":870,"completion_tokens":1851,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":486,"completion_tokens_details":{"reasoning_tokens":1773}},"tokens_in":486,"tokens_out":1851,"duration_ms":14774,"temperature":1.0,"reasoning_tokens":1773,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T13:07:46.203915+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Test Lemma 2.1.13 on a distributor that is $K$-flat but not cover-distributing, for example one induced by a flat but not cover-preserving functor: if $a_K \\mathrm{lext}_H i_J$ is not lex, then the lemma as stated is false. To test Theorem 2.2.19, look for two continuous distributors between the same sites that induce isomorphic geometric morphisms but are not themselves isomorphic; the theorem predicts none exist, since the equivalence is categorical.","supporting_citations":[{"cited_title":"Distributors at work","cited_arxiv_id":null,"evidence_quote":"Supplies the notion and basic theory of flat distributors, the foundation on which the paper's site-relative flatness is built."},{"cited_title":"Springer Science & Business Media, 2012","cited_arxiv_id":null,"evidence_quote":"Provides the flat-functor correspondence and filteredness criteria used to link flat distributors to lex inverse images and to extract distributors from geometric morphisms."},{"cited_title":"Algebraic theories in toposes","cited_arxiv_id":null,"evidence_quote":"Contributes the bidense-morphism continuity condition used to characterize distributors of sites in Proposition 2.2.9."},{"cited_title":"Morphisms and comorphisms of sites I -- Double categories of sites","cited_arxiv_id":"2505.08766","evidence_quote":"The companion first part of this program, whose double-categorical framework for morphisms and comorphisms the present distributor notion is designed to subsume."},{"cited_title":"Proarrows and cofibrations","cited_arxiv_id":null,"evidence_quote":"Supplies the definition of proarrow equipment used to discuss the equipment-like properties of the bicategory of sites and distributors."}],"review_version":1}