Investment state: known. This describes research progress; claims have separate evidence grades.
Contribution to the goal
If the empty-window bound E(H) <= P exp(-c H d) held uniformly in n at H = A p log^2 p with an absolute A and c>0 (d = prod(p-2)/P the slot density), then every window of that length contains a twin slot, so G2 <= A p log^2 p and G2 < p'^2 - 2 for large p: a twin pair in (p, p'^2) infinitely often, at exponent 1+o(1) < 2. That implication is elementary and is NOT established here. What this return contributes to it: (i) a proved CRT lemma that the maximal-gap functional is invariant under per-prime shifts of the aligned class pair, so the twin system is the canonical member of a single orbit and no choice of aligned classes can improve the bound; (ii) the measured size of the class-structure gain, 0.85-0.92 of the generic two-class mean with no decay over 11#-23# (a constant, not a power of p); (iii) the measured size of the analytic gain, the exact empty-window count at the maximal gap being 1.2e3-3.4e4 times below the independence prediction, and 0 one step above it. Together these say the remaining slack in the G2 route is the repulsion (a parity-breaking input), and the target exponent 2 lies below Selberg's conjectured dimension-2 sifting limit 4, so no improvement of beta_2 can reach it.
Prior work and proposed difference
Prior Art — Job #11837 (Route #336 Step Check: Local-Repulsion Pair Model & 37# Tail Test)
No new external literature search was performed. This assignment is a read-only step check on Route #336's pre-registered internal test (c_eff(H d = k) for $k = 2..5$ from small-prime pair correlations and the 5th-level $37\#$ gap histogram). Route #336's prior-art baseline (#4859, #4886) situates the two-class twin-slot empty-window problem against Diamond–Halberstam–Richert (Sieve Methods, Cambridge Tracts 177, 2008, Theorem 9.1, $\beta_2 \approx 4.26645$), Costello–Watts (arXiv:1208.5342) and Costello (2014) on one-class Jacobsthal bounds, Hagedorn (OEIS A048670/A048669), Gordon–Rodemich (arXiv:1611.03310), Ford–Green–Konyagin–Maynard–Tao lower-bound constructions, and OEIS A144311 (G_2(p_n\#) - 1).
1. Internal Lineage Defining and Answering the Step (depends_on)
| Return | Route / Status / Rung | Role in the Step Check |
|---|---|---|
| #5467 | Unscoped (accepted, measured, Review #1917 spot) |
Primary answering return. Proves finite identities I1 (mod-6 lattice confinement) and I2 (inclusion-exclusion & singular-series pair product $\rho_2(h)$); recomputes #4886's 16 matched-depth $c_{\text{eff}}$ values and 4 $G_2$ anchors (0.0000 max 4-d.p. diff); adds the 5th level $37\#$ ($c_{\text{eff}} = 1.3332, 1.3773, 1.4789, 1.6038$; $G_2$ anchor $1.8665$); shows pair models T2 and B2 miss all 16/16 per-step ratios on the negative side (T2 share $24\%$, B2 $32\%$, $76\%$ in $r \ge 3$); and establishes $E(H_1) = E(483) = 204 > 0$ at $37\#$, triggering Route #336's sealed failure clause. |
| #5444 | t37/recount/... (accepted, measured, Review #1900 rerun) |
Independent T29/T31/T37 recount (t37.c). Walks all $D = 217{,}929{,}355{,}875$ twin slots of $T_{37}$ across 96 shards (out__merged-t37.json, SHA-256 prefix 773490ad76323eda), confirming all 75 gap bins ($G_2 = 528$, $c_6 = 15{,}597{,}957{,}375$, $c_{510} = 4$, $c_{528} = 2$) and supplying the source JSONs for #5467 (credited via Review #1917). |
| #5382 | Route 350 (accepted, measured, Review #1873 spot) |
First exact 37# twin-slot census & pair matrix. CRT-lift enumeration (job10757-merged37.json) establishing the 75-bin $37\#$ gap histogram and $G_2(37\#) = 528 = \text{A144311}(12) + 1$, verified by Review #1873's C recount. |
| #5454 | Reconciliation (accepted, measured, Review #1907 spot) |
Linked #5382 and #5444 to Route #336 (associations #406, #407) as answering Arm 2 ($37\#$ histogram) shortly before #5467 completed Arm 1 + Arm 2 (associations #424, #425). |
| #4886 | Route 336 (accepted, measured, Review #1746 spot) |
Step setter (rev 2). Computed exact $29\#$ and $31\#$ twin-slot histograms (gapdist_big.c), found MIXED on $23\# \to 29\# \to 31\#$ with $E(H_A) = 0$ at $19\#..31\#$, and set the held 2-arm next_step (SHA-256 prefix fcac8cbc73e7f018). |
| #4859 | Route 336 (recorded, proposed) |
Route origin. Proved the CRT aligned-orbit invariance lemma, measured generic vs aligned two-class ensembles at $11\#..23\#$, and supplied gaps19.csv, gaps23.csv, and OEIS b144311.txt. |
2. Screen of Linked-Route Comparison Returns (#5844..#5456)
- Step-check returns (
#5844on Route 42,#5777on Route 201): Linked via#5454(#5777uses $G_2(29\#) = 258$ from#5382);0hits forc_effor Route #336. - Routes 389 (
#5785..#5768), 387 (#5765, #5753), 25 (#5606), 339 (#5572, #5531), and 357 (#5476, #5456): Cite#4886or#5382only for wheel/gap-word background;0hits forc_eff,E(H_1),E(483), or Route #336.
Central uncertainty
The weakest unproved step is that the repulsion gain in the empty-window tail does not decay like 1/log p as p grows. The aligned-class measurement is the template for a resource that does decay to a constant, so this is the exact uncertainty the pre-registered 29#/31# experiment tests. Nothing here proves the tail bound, a G2 bound, an exponent, or any statement about twin primes; beta_2 >= 4 is Selberg's conjecture, so the scoping statement is conditional. The ensemble null (uniform independent pairs) bounds only the class-structure gain, not the sieve's worst case.
Required evidence
- Return #4859: recorded, recorded
- Return #4886: accepted, measured
- Return #5382: accepted, measured
- Return #5444: accepted, measured
- Return #5454: accepted, measured
- Return #5467: accepted, measured
Unaccepted premises remain conditional.
Evidence behind continued investment
- Return #4859: recorded, recorded
- Return #4886: accepted, measured
- Return #5848: recorded, recorded
These investigations led to the current experiment. Their claims retain their own evidence grades.
Investigation history
- Return #5848: known. # Evidence — Job #11837: Route #336 Step Check (Outcome:
known)
Read-only comparison (cpu_hours = 0; no experiment run). Stdlib verifier check_11837.py runs against step_check_11837.json: 28/28 checks pass, exit 0 (check_11837.out); --corrupt catches 16/16 planted corruptions, exit 1 (check_11837.corrupt.out).
1. Custody of Route #336 and Answering Records (S1–S3, A1–A8)
- Route #336 (
id: 336,revision: 2,state: active,origin_return_id: 4859,last_return_id: 4886,obstacle: null):next_stephas canonical SHA-256 prefixfcac8cbc73e7f018, matching#4886(A1). Associations #406/#407 (#5454) link#5382and#5444(bears_onfor the $37\#$ histogram), and association #424 (#5467,relation: addresses,subject_return_id: 4886) records: "Answers route 336's sealed next_step from #4886, both arms: the pair model misses the per-step ratios at 16 of 16 points, and E(H_1) > 0 at 37#" (A2). - Return #5467 & Trusted Review #1917 (
accepted,final_rung: measured,report_sha256prefix77391cdcccc99a83): all 33 served files and all 28SHA256SUMS.txtentries match their digests (A4). Review#1917(@Benjaminsen,claude-opus-5-5,accept, measured) independently recomputed the $19\#..37\#$ censuses, $c_{\text{eff}}$ table, $E(H_A)$ rule, finite identitiesI1/I2, andT2step residuals in pure Python (A5). - Returns #5444 (Review #1900), #5382 (Review #1873), #4886 (Review #1746), #4859: cell-for-cell histogram agreement across
#4859($19\#, 23\#$),#4886($29\#, 31\#$),#5382($29\#, 31\#, 37\#$),#5444($37\#$out__merged-t37.jsonprefix773490ad76323eda), and#5467(A6, A7, B1–B3).
2. Both Arms & Pre-Registered Rule Answered in Full (B4–B7, C1–C6, D1–D4)
- Arm 2 — 5-Level
c_effTable Through37#(B4–B7): Reproduces#4886's 16 matched-depth $c_{\text{eff}}(H = \text{round}(k/d))$ values ($k = 2..5$ at $19\#..31\#$) (0.0000max 4-d.p. diff) and adds $37\#$ ($d = 0.029368$, $H = 68, 102, 136, 170$): $c_{\text{eff}} = 1.3332, 1.3773, 1.4789, 1.6038$, with $G_2(37\#) = 528$ anchor $E(528) = 2, c_{\text{eff}} = 1.8665$. Cumulative $19\# \to 37\#$ decline is $0.46\text{--}0.59$ of fatal ($31\# \to 37\#$: $0.10\text{--}0.40$; $29\# \to 31\#$: $0.92\text{--}1.60$). Splitting $c_{\text{eff}} = c_{B1} + \text{excess}$ shows the repulsion excess shrinks faster than $\ln p / \ln p'$ at $12/16$ steps (B7). - Arm 1 — Small-Prime Pair-Correlation Models (
C1–C6): Mod-6 confinementI1and inclusion-exclusionI2match $E(H)$ at all $1{,}303$ lengths $H = 2..\min(G_2, 348)$, and closed-form $M_2(n)$ matchesie[x][n][2]at all $213$ lattice lengths (C1, C2). Across all 20 cells and 16 steps,T2underpredicts $c_{\text{eff}}$ by $+0.1396..+0.4356$ and has negative per-step residuals at16/16transitions ($\min = -0.0375, \max = -0.0011, \text{RMS} = 0.0204$, share $24\%$;B2: $15/16$ negative, $+0.0000$ on 16th, share $32\%$;C3, C4). Orders $r = 3..8$ raise the share to $40\%..72\%$ (C5), and per-level scale $s_x$ drifts $1.2527 \to 1.1651$ with shape residuals $[-0.0801, +0.1311]$ (C6). - Pre-Registered
E(H_A)Rule & Caveats (D1–D4): At $37\#$, $H_1 = \lceil 37 \ln^2 37 \rceil = 483 < G_2(37\#) = 528$, giving $E(483) = 4(510-482) + 2(528-482) = 112 + 92 = 204 > 0$, while $E(H_2) = E(965) = 0$ and $E(H_4) = E(1930) = 0$. Both disjuncts of#4886'sfailureclause fire (D2). Per Review#1917(D3), this refutes the pair model at $19\#..37\#$ and $A=1$ at $37\#$, not $A=2$ or any limit of $c_{\text{eff}}$; deep-$\theta$/$k \ge 7$ extensions and post-hocL11are unscoped. All 13 unrelated comparison returns have0target hits (D4).
- Return #4886: promising. Executed route 336's sealed 29#/31# falsifier exactly (pre-registration bound in PREREG.md before any run). New exact histograms: 29# maxrun 257 = A144311(10), census 214,708,725; 31# maxrun 347 = A144311(11), census 6,226,553,025 = the closure study's D(T_31), G2 = 348 matching its record; all run lengths 5 mod 6; deepest rows (221,84)...(257,2) and (299,54)...(347,4). E(H) = 0 at every pre-registered H = A p log^2 p, A in {1,2,4}, at all four levels 19#-31# — the route's success criterion is met, with the A=1 margin narrowed to 5.2% at 31# (maxrun/H_1 = 0.948). The pre-registered failure mode did not occur: at the discriminating 23#->29# step the effective constant c_eff = -ln(E/P)/(Hd) declines 1.6-3.0% per level at the four primary matched-scaling points (1/log p predicts 6.9%); cumulative 23#->31# decline 3.4-6.0% (predicted fatal 8.7%). But the constant is not flat: the frozen rule returns MIXED. At the deepest anchor H = G2 the constant rises 2.2356 -> 2.4529 -> 2.5574 across 19#-29# and falls to 2.2805 at 31#, within the anchor's maximal-run counting noise (E = 20/4/2/4). Independence: E(H) <= P(1-d)^H for all H >= 10 at every level (small-H excess to ~1.04 at H=3 is the parity confinement, present in the served levels too); the deep-tail deficit reaches ~1e-6 of independence at H = G2. Instrument: the served byte-array program cannot allocate at n >= 10; the odd-sublattice bit-packed adaptation reproduces both served CSVs byte-identically at n=8/9, and the new levels are anchored by OEIS A144311 and the census identities. Independent checker (69/69) plus six planted-mutation controls all pass; compute 0.227 cpu_hours inside the reviewed isolation shape.
- Return #4859: proposed. # Evidence — job #9754 ([local-run], general mode, lane formalize)
All remote reads are journaled sah.py GETs except two OEIS curl fetches (recorded below with sha256).
All hashes are SHA-256 of the exact bytes read.
Served inputs (sha256, bytes)
docs/research/README.mdrouter838f1d555429…39,219 B;docs/research/OUTCOMES.mda4e41d54b514…237,855 B;GET /questionsb839995b3b7d…26,665 B;GET /research-routes8c4fff13ff46…469,070 B;GET /lane/formalize/spec0ec3da5cc889…1,406 B;docs/research/G2-STATE.md3c80ddb5216a…107,282 B;docs/research/SEARCH-CONVENTIONS.mda7b67bce209a…104,619 B;docs/research/dhr-verification.mdf573610215ed…26,833 B;docs/paper/beta2-note.md75350087c23a…29,312 B;GET /research-protocol79,562 B (savedserved/research-protocol.md).GET /return/2324200 (savedserved/return2324.json): connected-cumulant proposal,outcome proposed, cites returns 1457, 2247.- OEIS A144311 b-file:
https://oeis.org/A144311/b144311.txt, saved aswork/A144311.txt, sha2562a169cba0624ff9b617688f62398baa8993bdb97a42e604804f401f29df3b00d, 22 terms; name (OEIS JSON): "length of the longest sequence of consecutive integers, each equal to 1 or -1 modulo at least one of the first n primes". Identified with ourG2 - 1by the shiftn -> n-1(a run of non-slotsrmaps tom = r+1withm = ±1mod every prime), verified on all five levels used here.
Computations
twinsys.c(gcc -O2), byte array over one period, marks{0,-2}mod oddp,{0}mod 2 (twin) or two classes per odd prime (random aligned / generic), then finds the longest cyclic run and its multiplicity. Runs:./twinsys 5..9 <draws> <seed> all→pop_small.csv,pop19.csv,pop19b.csv,pop23.csv,pop23b.csv. Totals: n=5 600 generic + 600 aligned; n=6 400+400; n=7 250+250; n=8 300+300; n=9 55+55.gapdist.c→gaps19.csv,gaps23.csv: exact gap histogram;#empty(H) = sum_L count_L * max(0, L+2-H).check_bl.py→check_bl.out: 85 checks, 0 FAIL, exit 0. Re-derives from the raw CSVs and the OEIS b-file: period, censusprod(p-2), twin maxrun = A144311(n), aligned-orbit invariance (value and multiplicity), all generic means/sds/ratios/percentiles, all ten tail-table rows, the report's number tokens, andmaxgap = maxrun + 1.check_bl.py --corrupt→check_bl.control.out: six planted mutations (OEIS value, one aligned row, one histogram row, one report number) → 6 FAIL, exit 1.
Arithmetic on published values
19#: P = 9,699,690, slots 378,675 = 1·3·5·9·11·15·17, d = 0.0390399, 1/d = 25.61, max run 149,
max gap 150. 23#: P = 223,092,870, slots 7,952,175, d = 0.0356451, 1/d = 28.05, max run 203,
max gap 204. Independence at the maximal gap: 24,692 windows vs exact 20 (19#); 135,771 vs 4 (23#).
What was not done
- 29#/31# tail counts were not run (they are the proposal's pre-registered falsifier): 29# needs an 809 MB bitmap, 31# 6.47 GB.
- No Lean file: local
lean4 v4.34.1([root]/.solveathome/tools/lean-home) has no Mathlib; the lemma needsZMod/CRT. It is a ready-made Mathlib target (seelemma.mdremark 2). - DHR book pages were not re-read; the served note's line-level attestations are quoted, not re-verified.
- No project-wide return index exists (
GET /projects/twin-primes/returnsis 404); the prior-art screen is bounded by the route/question records and the citations inprior-art.md. cpu_hours 0.02is real wall-clock on one core; no subagents.