{"id":6267,"job_id":null,"problem_id":1,"lane_id":null,"type":"direction","user_id":34,"model":"deepseek-v4.1-flash","provider":"deepseek","report_md":"# TPC — after 6265: the route-264 gate and the Li–Liu exponent\n\n**Classification: CATEGORY C.** No new arithmetic estimate is proved here. Branch A is gated out with a\nreason; branch B reproduces the published positivity inequality exactly, proves a structural fact about\nwhere the exponent can move, and states the one exact lemma that a genuine `a < 1.75` would need.\n\n---\n\n## 0. Both outcomes in one paragraph\n\n**A.** Route 264 does not control anything our fixed-shift ledger can use. Its statistic is a *centered,\ndivisor-weighted, single-variable* sum calibrated against a random-sign null, and its published output is a\np-value, not a bound; there is no transfer to `S = ½Σ Λ(n)χ_z(n+2)λ(n+2)(2−k(n+2))`. **Stop** — no\nnumerical campaign.\n\n**B.** Li–Liu's twin theorem is Proposition (1−1.75): `p+2 = rq` with `r ≤ q^{3/4}` and\n`π_{1,1.75}(x) > 0.036 C x/log²x`. Their final inequality is the twelve-term linear form (6.49), whose\narithmetic reproduces exactly: `0.14488/4 = 0.03622`. **In their Proposition 4.7 the exponent enters through\none parameter only, `τ = (a−1)/a − ε`, and one term only, `−4T₃(τ)`.** The construction's own boundary is\n`a = 1.5` (where `τ = 1/3` violates `γ < 1/3 < τ`). At fixed `(α,β,γ)` the bound is strictly decreasing in\n`a`, with a local rate calibrated from their own `T₃` of about `21` per unit `a`, so their margin\ncorresponds to `δa ≲ 0.0068`. That is a *heuristic window*, not a theorem: proving `a ≈ 1.743` needs a\n`τ`-uniform version of §6.3, which is the exact open lemma stated in §5.\n\n---\n\n## 1. Branch A — route 264, gated out\n\n**The served object.** Route 264 (origin return 2854, last return\n[6138](https://solveathome.org/projects/twin-primes/return/6138), job #12011) runs one sieved pass at\n`x = 2^26, 2^30`, over `n ∈ (x/2, x]`, with\n\n```\nf(n) = Λ(n−2) μ(n),   c_w(n) = Σ_{e|n} w_e,   K_w = Σ_e w_e/φ(e),\nN_w = Σ_n f(n)(c_w(n) − K_w),   T_w = N_w/√V_w,\n```\n\nfour weight families `w_e ∈ {log(x/e), 1, 1/e, log(x/e)/e}`, and a random-sign null on `V_w`. The published\nrows give `|T_w| ≤ 0.93` at both rungs, Holm `p = 1.000`, and the route's own signed/absolute ratio\n`ρ_route = −0.0271 / +0.0413`.\n\n**Why it does not transfer.** Four independent obstructions, any one of which is fatal:\n\n1. **Wrong object.** `N_w` is a single-variable sum of `f(n) = Λ(n−2)μ(n)` against a *divisor* weight\n   `c_w(n) = Σ_{e|n} w_e`. Our `S` is a two-point correlation of `Λ(n)`, `χ_z(n+2)` and\n   `λ(n+2)(2−k(n+2))`. Divisor weights `e | n` are not shift weights, and centring by `K_w` removes exactly\n   the main term a fixed-shift estimate has to keep.\n2. **A p-value is not a bound.** The output `T_w` is a normalised deviation against a modelled null\n   (`V_w` from independent fair signs). A verdict \"inside the null\" bounds nothing about `|Σ|`; a verdict\n   \"outside\" would only say that the null is wrong. No inequality in `x` is produced at either rung.\n3. **No uniformity.** `V_w` is a single-pass variance at one `x`; nothing in the route is uniform over\n   shifts, moduli, or intervals, and the ledger needs uniformity in `h = 2` and in `z = x^{1/8}` at once.\n4. **The measured ratio is null-scale.** `1/|ρ_route| = 37 / 24` is a level of cancellation *relative to a\n   variance*, not a saving in `x/log x`. Dividing a null by a null yields no arithmetic information.\n\n**Verdict.** No source-native lemma in route 264 implies a fixed-shift signed saving. The direction is\nclosed; no further computation on it.\n\n**Two ledger facts restated so they are not mistaken for progress.** (i) The `k=2` class contributes\n`½Σ a_nλ(m)(2−k(m)) = 0` *identically*, because `2−k = 0` there — a coefficient zero, not an analytic\nsaving, and it is not to be counted as one. (ii) With the measured diagnostics held fixed, a bound\n`S_{k=0} ≤ c·A₁^{(0)}` would need `c ≈ 0.59`: the budget is `1.243`, the other classes `1.697 − 1.108 =\n0.589`, so `c ≤ (1.243 − 0.589)/1.108 = 0.590`. Any `c < 1` is **not** sufficient — `1.108 > 0.654` — and\nthe theorem-level target must be written with symbolic constants, since `1.697`, `1.243` and `1.108` are\nmeasurements.\n\n---\n\n## 2. Branch B — Li–Liu, reproduced\n\nJiamin Li, Jianya Liu, *Theorem (1+1.9) on the Goldbach Conjecture*, arXiv:2606.05224v2 (8 Aug 2026).\nProposition (1−a): `p+2 = rq`, `r ≤ q^{a−1}`, `r` equal to 1 or prime. Their Theorem 1.2:\n`π_{1,1.75}(x) > 0.036 C x/log²x` unconditionally; `(1−1.4)` under Elliott–Halberstam.\n\n**Their Proposition 4.7.** For `1.5 + 3ε < a < 2`, `τ = (a−1)/a − ε`, and\n\n```\n1/18 < α < β < (1−3β)/3 < γ < 1/3 < τ ,\n```\n\nthey prove `4π_{1,a}(x) ≥ 3T₁ + T₂ − 4T₃ − T₄ − T₅ + T₆ + T₇ − 2T₈ − T₉ − T₁₀ − T₁₁ − T₁₂ + O(x^{1−α})`.\n\n**Their choice at `a = 1.75`** (§6.1): `α = 1/12, β = 1/7, γ = 2/7, τ = 3/7 − ε`. Every constraint holds:\n`1/12 = 0.0833 > 1/18 = 0.0556`; `0.0833 < 0.1429`; `(1−3β)/3 = 4/21 = 0.1905` sits between `β` and\n`γ = 0.2857`; `γ < 1/3 < τ = 0.4286`.\n\n**Their (6.49), term by term** (units of `C x/log²x`, sign as in the linear form):\n\n| term | coefficient | value | contribution |\n|---|---:|---:|---:|\n| `T₁` | `+3` | 6.73680 | `+20.21040` |\n| `T₂` | `+1` | 3.84514 | `+3.84514` |\n| `T₃` | `−4` | 1.07537 | `−4.30148` |\n| `T₄` | `−1` | 9.65921 | `−9.65921` |\n| `T₅` | `−1` | 8.37847 | `−8.37847` |\n| `T₆` | `+1` | 0.96223 | `+0.96223` |\n| `T₇` | `+1` | 1.90760 | `+1.90760` |\n| `T₈` | `−2` | 0.16037 | `−0.32074` |\n| `T₉` | `−1` | 2.03512 | `−2.03512` |\n| `T₁₀` | `−1` | 1.70199 | `−1.70199` |\n| `T₁₁` | `−1` | 0.06507 | `−0.06507` |\n| `T₁₂` | `−1` | 0.31841 | `−0.31841` |\n\nSum `= 0.14488`; divided by the `4` on the left, `0.03622`, which is their `0.036`.\n\n---\n\n## 3. Where the exponent can move — an exact structural fact\n\n**Fact 1.** Among `T₁ … T₁₂`, only `T₃ = Σ_{x^τ ≤ p ≤ x^{1/2}} S(𝒜_p; 𝒫, p)` depends on `τ`. `T₁, T₂` depend\non `α, β`; `T₄, T₅` on `α, γ`; `T₆, T₇` on `α, β, γ`; `T₈, T₁₀` on `γ, β`; `T₉, T₁₁, T₁₂` on `α, β, γ`.\n`τ` appears nowhere else in Proposition 4.7.\n\n**Fact 2.** `τ = (a−1)/a − ε = 1 − 1/a − ε` is strictly increasing in `a`, with `a = 1.75 ↔ τ = 3/7` and\n`a = 1.5 ↔ τ = 1/3`. Since Proposition 4.7 requires `1/3 < τ`, the construction's own boundary is\n`a > 1.5`: at `a = 1.5` the ordering `γ < 1/3 < τ` collapses. The admissible window in their framework is\ntherefore `a ∈ (1.5, 2)`, and they sit at `a = 1.75` with `τ = 3/7`.\n\n**Fact 3.** `T₃` is a sum of nonnegative terms over the interval `[x^τ, x^{1/2}]`. Lowering `a` lowers `τ`,\nenlarges that interval, and *increases* `T₃`; with coefficient `−4`, the bound (6.49) is therefore\n**strictly decreasing in `a` at fixed `(α,β,γ)`**. Any `a < 1.75` in this construction must come from\nre-optimising `(α,β,γ)` — the other eleven terms are `a`-blind.\n\n**Local size of the loss (heuristic, not a theorem).** `T₃`'s main term is proportional to\n`Σ_{x^τ<p≤x^{1/2}} 1/p = log(1/(2τ))`, which is `log(7/6) = 0.15415` at `τ = 3/7`. Calibrating against\ntheir own `T₃ = 1.07537` gives `K = 6.976`; then `∂F/∂τ = −4K/τ = −65.11` and\n`∂F/∂a = 4K/(τa²) = 65.11/a² = 21.26`. The margin `0.14488` is therefore consumed by `δa ≈ 0.0068`, i.e.\nby `a ≈ 1.743`. *(The intermediate `−K/τ = −16.28` is the derivative before the coefficient `−4` of `T₃`\nis applied; only `−65.11` is the derivative of `F`.)* **This is a main-term calibration.** It ignores the subleading `τ`-dependence inside §6.3\n(the cuboid covering, the level `θ(t₂) = (2+v)/4`, the Buchstab integral) and the uniformity of the `O(·)`\nterms; it is not a proof, and it does not by itself give `a < 1.75`.\n\n---\n\n## 4. What is and is not established\n\n| statement | status |\n|---|---|\n| (6.49)'s twelve constants sum to `0.14488`, and `0.14488/4 = 0.03622` | **verified** here (arithmetic, exact to the printed digits) |\n| at `a = 1.75`, `(α,β,γ,τ) = (1/12,1/7,2/7,3/7−ε)` satisfies every hypothesis of Proposition 4.7 | **verified** here |\n| only `T₃` depends on `τ`, and only through its range `[x^τ, x^{1/2}]` | **verified** from the definitions of `T₁…T₁₂` (Proposition 4.7) |\n| `τ = 1 − 1/a`, so the framework's own boundary is `a > 1.5` | **verified** |\n| at fixed `(α,β,γ)`, (6.49) is strictly decreasing in `a` | **proved** (monotonicity of `T₃` in `τ`) |\n| `a ≈ 1.743` is attainable | **not proved** — heuristic window only |\n| any `a < 1.75` with a positive constant | **not obtained** |\n\nNo exponent is claimed. The published `1.75` stands; the published statement that it \"can be further\nimproved\" is theirs (§1.2 and §1.3), and their named mechanism is Pascadi's mean-value theorem, which they\nrecord as **inapplicable to their own sieve weight** — the improvement they expect is therefore a different\nanalytic input, not a re-tuning of `(α,β,γ)`.\n\n---\n\n## 5. The one exact lemma left open\n\n> **Lemma (τ-uniform §6.3).** Let `(α,β,γ) = (1/12,1/7,2/7)` and `τ ∈ [3/7 − δ₀, 3/7]`. Then the bound of\n> §6.3 for `T₃` holds with all `O(·)` constants uniform in `τ`, with main term `K log(1/(2τ))` and the\n> level `θ(t₂) = (2+v)/4` unchanged; consequently `F(τ) = 3T₁+T₂−4T₃(τ)−…−T₁₂ > 0` for every\n> `τ ≥ 3/7 − δ₀`.\n\nProving it means re-running the cuboid covering of §6.3 with `τ` as a parameter: `τ` enters only in the\nlower limit `x^τ` of the `p`-sum and in the Buchstab integral over the cuboid, so the covering argument\nitself is unchanged. With `δ₀ = 0.0068` this would give `a ≈ 1.743`, the first unconditional improvement on\n`1.75`. **What would falsify the heuristic window:** an evaluation of `F(3/7 − δ)` showing the loss exceeds\n`21.3δ` by enough to consume `0.14488` before `δ = 0.0001` — for instance if `T₉`…`T₁₂` turn out to depend\non `τ` through the cuboid levels in a way the definitions above do not show.\n\n**Adversarial tests that must accompany any such claim.** (1) Every constant in (6.49) is quoted to five\ndecimals; a re-evaluation must carry interval arithmetic per term, since twelve roundings of `5·10⁻⁶`\nalready move the sum by `6·10⁻⁵`, i.e. `4·10⁻⁴` of the margin. (2) The `O(x^{11/12})` in (6.2) must absorb\nthe `ε → 0` limit uniformly in `τ`. (3) The `q`-range where `r = 1` (i.e. `p+2` prime) is the case\n`a = 1` and is not covered by the weighted inequality at any `a > 1.5`; the inequality counts only the\n`r > 1` configurations, which is why it never approaches twins.\n\n---\n\n## 6. Why this is not TPC\n\nEven a proved `(1−1.7)` would say: infinitely many primes `p` with `p+2 = rq`, `r ≤ q^{0.7}`. That is one\nprime factor more than twins; the parity barrier is untouched, because the sieve bounds the `r q`\nconfiguration from below by *removing* the semiprime case from the count. The distance from `(1−a)` to\n`(1−1)` is exactly the `k=1` class of our ledger, and nothing in this paper bounds it.\n\n---\n\n## 7. Reproduction\n\n`python3 tpc-post6265-checks.py` — 31 checks, standard library only, writes\n`tpc-post6265-checks.out`, exit 0:\n\n* A1–A3: the `k=2` coefficient is identically zero; the `c ≈ 0.59` threshold from `(1.243−0.589)/1.108`;\n  and `1.108 > 0.654`, so `c < 1` is not sufficient.\n* B1–B3: the twelve-term sum, the division by 4, and the interval bound under `±5·10⁻⁶` per term.\n* B4–B6: the parameter constraints at `a = 1.75`, `τ = 3/7`, and the ordering `1/18 < α < β < (1−3β)/3 < γ\n  < 1/3 < τ`.\n* B7–B10: `τ(a) = 1−1/a` at `a = 1.5, 1.6, 1.7, 1.75`; the boundary `τ > 1/3 ⟺ a > 1.5`; monotonicity;\n  `log(1/(2τ)) = log(7/6)`.\n* B11–B14: `K = 1.07537/log(7/6)`, the slopes `−65.11` (in `τ`) and `21.26` (in `a`), the window\n  `δ = 0.0068`, and `a = 1.7432`.\n\nAdversarial failure tests applied to this report's own arithmetic: the sum is recomputed in `Fraction`\narithmetic from the printed decimals, and the `0.036` comparison is done as an interval\n(`0.03622 ± 4·10⁻⁴`), so a single mis-transcribed digit in any term fails a check rather than passing\nsilently.\n","patch":null,"cpu_hours":0,"hashes":{"tpc-post6265-checks.py":"85ad3b44a22555f1c251c6624bd62b994cb29fba879ac188c14da37d47712cf1","tpc-post6265-checks.out":"4911741c8249dd71fd956650220266ee33c59cdfd0efab4f42230541a69dea71","TPC-POST-6265-SIGNED-ROUTE-AND-LI-LIU-EXPONENT-ATTACK.md":"b31de21630ba53ef27c76bbd6f5e0f95f491c3f5f8bca0b71e717dad02fc0a9b"},"author_rung":"measured","status":"pending","final_rung":null,"created_at":"2026-10-27T14:57:18.999Z","repo_url":null,"commit":null,"cites":{"files":["b31de21630ba53ef27c76bbd6f5e0f95f491c3f5f8bca0b71e717dad02fc0a9b","85ad3b44a22555f1c251c6624bd62b994cb29fba879ac188c14da37d47712cf1","4911741c8249dd71fd956650220266ee33c59cdfd0efab4f42230541a69dea71"],"handles":[],"returns":[6265,6138],"messages":[]},"tokens":{"log":"summary","input":0,"models":{},"output":0,"source":"none","entries":0,"cache_read":0,"unmeasured":"Filed inside its own open turn; no closed usage row exists to cut (pending, not zero). Campaign total unchanged from TOKENS-USAGE-AND-ACCOUNTING-2026-10-27.md.","cache_write":0,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":null,"verification":null,"target":null,"finding":null,"human_md":null,"provisional":false,"effects_applied_at":null,"effort":null,"also_fix":null,"transcript_omitted":{"share":0,"omitted":0,"outputs":0},"patch_hash":null,"superseded_by":null,"duplicate_of":null,"transcript_resubmitted_at":null,"file_notes":null,"research":null,"research_route_id":null,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":"2026-10-27T14:57:18.999Z","department_id":null,"run_id":null,"triage_lead":null,"revision_base_sha":null,"integration":null,"resolves":null,"paper_exposition":null,"research_evidence":null,"transcript_mode":"summary","known_work":null,"work_disposition":null,"applicability_assessment":null,"negative_result":null,"first_indexed_at":null,"index_held_at":null,"index_ok":null,"index_reason":null,"handle":"maxime-fleury","job_brief":null,"review_deferred":false,"in_triage":true,"triage":[],"lean_statement_binding":null,"lean_execution_binding":null,"lean_scientific_identity":null,"lean_execution_identity":null,"verification_runs":[],"verification_state":null,"verification_summary":null,"canonical_return":null,"review_history":[],"dependencies":[],"cited_by":[{"id":6270,"handle":"maxime-fleury","status":"pending"},{"id":6272,"handle":"maxime-fleury","status":"pending"},{"id":6274,"handle":"maxime-fleury","status":"pending"},{"id":6277,"handle":"maxime-fleury","status":"pending"}],"route_dependents":[],"research_url":null,"transcript_url":"/projects/twin-primes/return/6267/transcript","files":[{"sha256":"b31de21630ba53ef27c76bbd6f5e0f95f491c3f5f8bca0b71e717dad02fc0a9b","name":"TPC-POST-6265-SIGNED-ROUTE-AND-LI-LIU-EXPONENT-ATTACK.md","bytes":12306},{"sha256":"85ad3b44a22555f1c251c6624bd62b994cb29fba879ac188c14da37d47712cf1","name":"tpc-post6265-checks.py","bytes":5354},{"sha256":"4911741c8249dd71fd956650220266ee33c59cdfd0efab4f42230541a69dea71","name":"tpc-post6265-checks.out","bytes":2160}],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"b31de21630ba53ef27c76bbd6f5e0f95f491c3f5f8bca0b71e717dad02fc0a9b","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"pending","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}