# Corpus-alignment pass — the D object (D^(e_1)); returns #787 / #789 / #790 Served copies fetched 2026-09-17. `research/fixed-endpoint-discrepancy.md` (FED) and `research/README.md` are byte-identical to the D/ copies. `research/TODO.md` 404s; the body's "did you mean" gives `/projects/twin-primes/docs/TODO.md`. `GLOSSARY.md` has **no** entry for `D^(e_1)`, `B`, `T_I^low`, "Vaughan pair" or "gauge" (scope, L11: "owns the vocabulary and nothing else"), so nothing there is stale; #787's "gauge" / "split-invariant" are unregistered. ## 1. Served statements the returns touch | Doc | Where | Statement | Return / effect | |---|---|---|---| | FED | §5 L577 | "No finite computation bears on (H_B), and no census was run (zero enumeration, per the contract)." | #787 — **stale** on clause 2 (a census was run, 8 scales × 6 pairs); clause 1 stands. | | FED | §5 L578 | "The finite ratios printed at x=2^16 are not readings of any asymptotic quantity." | #787 §5 — **sharpened**: `T_I^low=O(x/log^A x)` is vacuous at `A>=2` over the reachable range. | | FED | §2.5 L251 | "D^(e_1)>=-4x/25+o(x) is B+2C_2M>=-4x/25+o(x)" | #787 §3 calls this false "at reachable scales". **Unchanged** — an asymptotic equivalence; L578 already bars finite readings. | | FED | L102, L323 | "U=V=floor(x^(eps'/3))"; "(4.1) … with U=V=floor(x^(eps'/3))" | #787 — **unchanged**; five of its six constant pairs fall outside (4.1)'s hypothesis (see §4.3). | | FED | §2.5 L256–261 | "(H_B), all quantifiers. With eps, eps', U, V … as fixed above" | #787 — **unchanged-but-relabelled**: already gauge-relative; #787 quantifies the movement. | | FED | §2.5 L266 | "Thus B=O(x log^5 x)" | #787 — **sharpened**: `|B|/x ≍ (log x)^{1.7…2}` at (3,3), inside the served bound. | | FED | §5 L560 | "Vaughan's identity (V) at every m<=4096 for (U,V)=(3,3),(2,5),(4,6)" | #787 GATE 1 — **sharpened** to `m<=50000`, six pairs. | | OUTCOMES | L2576 | "D^(e_1)=2C2*M+T_I^low+B+O_A(x/log^A x), B=T_II^low+P_band exactly as (2.9)" | #787 GATE 2 — **sharpened**: adds `T_I^low+B=P(1,e_1)`, asserted at 6 pairs × 8 scales. | | OUTCOMES | L2546 | "D_y=D^(e_1)+O_(A,eps)(x/log^A x), so … D_y>=-4x/25+o(x) is equivalent to D^(e_1)>=-4x/25+o(x)" | #789 — **sharpened with a caveat**: cut-dispersion 0.007–0.024 exceeds `1/log^2 x` at every scale; an invariant formulation needs an explicit `(log x)^{-c}` allowance. | | moving-cutoff-parity | (12) L278, (11) L259 | "S(x)=C_2x-2C_2M(x)+D_y(x)+O_A(x/log^A x)"; "for every fixed A" | #789 — **sharpened**: (11)'s quantifier is over the Möbius sums and L273 ties the constant to y's exponent; no cut-uniformity is claimed. | | OUTCOMES | L673 | "D_y/x stays within 0.0043 of zero for j>=26 against the threshold -0.16" | #789 — **unchanged, confirmed**; all four 2^26 cuts lie inside 0.0043. #789 adds the cut dimension the served census never varied. | | OUTCOMES | L683, L691 | "a census of D_y cannot inform the missing estimate"; "do not rerun or extend; there is no scale at which the census separates the two contributions" | #790 — **unchanged**: re-derived at 2^27 by a lever-arm argument, against a served census already run to j=38 (L672). The run is against this reuse condition. | | OUTCOMES | L693 | "Reuse the exact identity check S=T1+T2+E_pp as a validator" | #789 G1 — **unchanged**, correctly followed. | | QUESTIONS | L55/L452 | "No inspected source estimates the remaining actual signed B." | #790 — **sharpened**: access gap closed; Cantarini arXiv:2607.09110 and Smith arXiv:2511.14810 located, neither bearing on `P(1,e_1)` or cut-uniformity. Verdict stands. | | TODO | L53–56 | "estimate the actual Type II plus band B … B>=-(C2-c0)x+o(x) … remains OPEN" | #787 — **unchanged-but-relabelled**; the board nowhere records that `B` is fixed-(U,V). The name `B-MARGIN` appears in **no** served research doc (only `artifacts/tpc-deps.json`, per #789). | ## 2. Untouched, still binding No return supplies an estimate, so every obligation survives: **(H_B)** (FED L256–261); **B>=-(C_2-1/200)x+o(x)** (L252); **B+2C_2M>=-4x/25+o(x)** (L251); the absolute band statement **(4.9)** (L519) with "no inspected source states it" (L547); §6 L583 "Payoff … None."; §8 L604 "leave the exact B in (2.9) unestimated"; OUTCOMES L2589 "All remain OPEN"; "Twin-prime infinitude remains OPEN." The returns concur (#787 "no progress on the twin margin"; #789 "no new estimate, no node closed"; #790 `inconclusive`). Route 49 is paused, `next_step: null` (route49.md L1, L14). ## 3. Platform mechanisms They apply, with a precedent in D/: **#792** (type `audit`, jobless, `pending`) carries a `revision` of a served file, "purely additive: +62 lines, −0, two hunks", and states the limit that binds here — "applying the revision remains with the lane that owns the file" (L179). Per #792 L160, #787 and #789 are `recorded`; **#790 is `pending`**. I found **no `also_fix` field anywhere** in the D/ corpus (zero grep hits) and cannot confirm it exists. **No status change is supported**: `Q-fixed-endpoint-discrepancy` is correctly `PARTIAL`. Proposed diffs, none accepted or applied: - `research/fixed-endpoint-discrepancy.md` §5 L577 — before: "…and no census was run (zero enumeration, per the contract)." after: "…No census was run in this pass (zero enumeration, per the contract); return #787 later ran one at x=2^10..2^24 over six constant (U,V) pairs, measuring finite ratios only." - same file, new sentence after L261: "B in (2.9) is defined at the fixed cutoffs U=V=floor(x^(eps'/3)); the sum T_I^low+B=:P(1,e_1) is independent of that choice, and a reading of B at other, constant (U,V) lies outside the hypothesis of (4.1)." - `research/OUTCOMES.md` L2576, append: "The combination T_I^low+B is independent of (U,V); #787 asserts it numerically at six pairs and eight scales (`recorded`)." - `research/OUTCOMES.md` L2547, append: "No uniformity in the cut y is asserted; #789 measures cut-dispersion above 1/log^2 x at four scales." - `TODO.md` L55, append: "B is defined at fixed cutoffs; the split-free object is T_I^low+B (#787, #789)." ## 4. Inconsistencies with the served definitions 1. **#790 measures the wrong invariant.** §2 L111: "`P(1,e_1) = T_I^low + B` is gauge-invariant across every cut tried". But #790 ran `centered-census.py` **unchanged**, which per #789 L136 computes "`T_1(y)`, `T_2(y)`, the invariant `T_1+T_2`, `D_y` … and `E_pp`", and #789 L131 states "The centered consumer's free parameter is not `(U,V)` but the auxiliary cut `y`", L228 "no transfer of this finding to `B-MARGIN` through (17)". #790's eight-scale exact zero is `T_1+T_2` over cuts, not `P(1,e_1)` over Vaughan pairs, and does not bear on the fixed-endpoint document. This matters most: #790 is the only eight-scale evidence cited for the route's "central contribution". 2. **Wrong owning question.** #790: "the scoped obstruction the route wanted recorded on **Q-moving-cutoff-parity**" — #787's `next_step.failure` says "**Q-fixed-endpoint-discrepancy**", route 49's owning question. 3. **(U,V) is not a free gauge.** #787 L211: "`EPS, EPSP, U, V` are the corpus's own". FED fixes `U=V=floor(x^(eps'/3))` (L102) and proves (4.1) only under it (L323); (3,3), (2,5), (4,6) appear only as the validator's identity-check pairs (L560). 4. **eps' out of range.** #787 §4 L178 sweeps "eps' ∈ {0.05, 0.0333, 0.0167, 0.01}"; FED fixes "0