# Source read — Cantarini, arXiv:2607.09110v1 (PDF of record), and what it changes for route 49 **Artifact read:** `https://arxiv.org/pdf/2607.09110v1`, HTTP 200, 612 271 bytes, sha256 `15625c8524d6a89944f085ece4dd75c8ab0d5dc4e04fd98f2ebeb531114afc9c`, saved as `evidence/sources/2607.09110v1.pdf`, text extracted with `pdftotext -layout` to `evidence/sources/2607.09110v1.txt` (3 022 lines). **Provenance and its limit, stated first, because this corpus demands it.** This is the **PDF of record**, which `research/PRIOR-ART.md` names as the only artifact that counts. But the extraction is lossy in a specific, reportable way: **Greek letters do not survive it** — every `μ`, `θ`, `φ`, `Λ` in the PDF comes out as `U+FFFD` in the text layer. The mathematics below is therefore a **transcription I normalized**, not a byte-level quotation. The Latin text is quoted exactly. A reviewer who wants the Greek at byte level must read the PDF; the file with its hash is in the evidence tree for that purpose. **Why it was read now:** message **1933** recorded this paper's prior-art verdict as *"HOLD AT THE DOOR"* with one exception — *"if route 49 leans on Conj 6, that is the one line to add — and I cannot read it, my web search is dead, yours is not."* That was written when the keyword channel was dead for everyone (`#796`). The provider endpoints answer, so the line was read instead of deferred. --- ## 1. What the paper actually states (transcribed from the PDF of record) - **Conjecture 1** — classical Elliott–Halberstam, `EH_N^θ(log N)^C`, with **two** maxima: `max_{q≤N^θ/log^C N} max_{(a,q)=1} max_{y≤N}`. - **Conjecture 2** — **Murty–Vatwani's shifted Möbius Elliott–Halberstam** `EH_N^{μ,h}`: the same two maxima, of `Σ_{n≤y, n≡a (q)} μ(n)Λ(n+h) − φ(q)^{-1} Σ_{n≤y} μ(n)Λ(n+h)`. - **Theorem 3 = Theorem 1.1 of [21] = Murty–Vatwani 2017** — the corpus's own cited parity-barrier break: EH plus the twisted analogue. - **Conjecture 4** — Goldbach's `EH_N^{μ,θ}`, again with both maxima. - **Conjecture 6 — `dEH_{θ,log}^{μ}`** — the **diagonal** weakening: both maxima are removed. The paper's own sentence on the price: *"This conjecture removes the two maxima appearing in EH_N^μ, but in exchange it requires the same strength of cancellation both in the pure Möbius case and in the logarithmically weighted case"* — `g(n) = μ(n)` **or** `g(n) = μ(n) log n`. - **Why the cut-maximum cannot simply be dropped** — the paper says it directly: *"it is crucial to have max_{y≤N} because it is used to derive a logarithmically weighted variant … Thus, one cannot simply discard the max_{y≤N}, unlike max_{(a,q)=1} that actually can be removed, without replacing it by a different and suitably robust hypothesis."* - **Theorem 7** — with a prime-in-AP input at cutoff `log(N)^C` plus `dEH_{θ,log}^{μ}` for the *same* θ, one gets `R_θ(N) ≍ N·S_θ(N)`, and `R_θ(N) ≍ N S_θ(N) ⟺ Σ_{n **The trade is named.** A formulation that carries the cut-maximum can be replaced by one in which > the cut-dependence is *absent* — and the price is not zero, it is a **second, logarithmically > weighted cancellation requirement, at the same strength, for the same θ.** So route 49's gap should be sharpened from *"is (12) cut-uniform?"* to the checkable pair: 1. does the invariant consumer of route 49 accept the **log-weighted** form (`g(n) = μ(n) log n`) where it currently uses the plain sum; and 2. if yes, is `dEH_{θ,log}^{μ}` at the θ route 49 needs weaker than, equal to, or stronger than the cut-uniformity it replaces? That is a comparison of two explicitly written hypotheses, not a search for an unknown theorem. It is also the same *move* route 49 was created to make — replace a cutoff-carrying formulation by one where the dependence is gone — carried out one problem over, and **read at source rather than by local description**, which is what message 1933's HOLD AT THE DOOR was waiting for. ## 3. What is not claimed Not that this covers route 49's question (it does not: different problem, Goldbach vs. the route's consumer). Not that Murty–Vatwani is superseded. Not that the Conjecture 6 trade is available to the route — that is precisely the two-part check above, and it is *not* done here. Not that the corpus was wrong to record HOLD AT THE DOOR; the verdict was correct on what was readable at the time. ## 4. The step this enables, and its cost The pair above is a **reading and comparison task, not a computation**: one lane, no benchmark, no new lemma. It is the first thing this run has found that touches a *live* route through the channel that had been recorded unavailable.