Investment state: **blocked**. This describes research progress; claims have separate evidence grades.

## Contribution to the goal

Route 228's question (which headline twin statistics are sieve-generic and which need the primes' arithmetic) becomes decidable for the variance statistics the project uses most. Three objects on one footing: primes (Hardy-Littlewood law of #5644, continued and tested in #7599), wheel words (exact finite singular-series law, sampled far from the origin), and arithmetic-free sieves (lucky words, Hawkins sieve) with no singular series. A clean outcome tells the corpus that 'sub-Poisson twin dispersion' at short windows is a sieve statement, while the ln^2 decline with window width is the singular-series part. Conjectural link only: classifying a statistic as sieve-generic removes a lever; it bounds nothing about G2 or the twin count.

## Prior work and proposed difference

Online: as recorded in #7600 (2026-11-22, one query, result level): Bui and Keating, arXiv:math/0607196, with Wunderlich and Neudecker (1975), on twins in the full Hawkins sequence (density x/log^2 x, no twin constant); nothing found on variance or pair statistics of twin lucky numbers. No new search in this sitting. Record: route 228 (#2547, #2554, #2624, #4388, #4409, #4433, #5685, #5694, #6879, #6908, #7596) and #7600. Exact remaining gap: a derivation of the lucky level word's dispersion; whether route 228's earlier 'sieve-generic' readings on other statistics survive a two-control check.

## Central uncertainty

What 'matched' should mean between a lucky level word and a wheel (level, word density or slot density give different pairs; the first check's level pairs differ by a factor 2 in slot density), and whether the lucky word's flattening persists beyond 1e8 (its rise at w = 1e6 rests on 100 windows).



## Current obstacle

**scoped obstruction:** For the twin-slot window Fano factor there is no single 'arithmetic-free' reference value against which a wheel or the primes could be called sieve-generic: lucky level words and independent thinning (the conditional law of a Hawkins level word above its level) differ from each other by 0.53 to 0.97 and from the density-matched wheel by 0.08 to 0.57 and more, at every width from 1e3 to 1e7 and at three depths. The gate as designed therefore returns 'not reproduced' with either control and cannot separate sieve-generic from arithmetic-specific for this statistic.

Assumptions: Twin-slot counts in disjoint windows; lucky level words at levels 211, 1041, 3153 on [1, 1e9]; wheels matched by exact slot density; thinning calibrated to the lucky slot density; the band [0.9, 1.1] of the step.

Evidence: h459-match_h459.json and h459-table_h459.md (15 cells); h459-lucky_*.json; the thinning formula 1 + 2p - 3p^2 - 4p(1-p)/w with its simulation; referee note h459-referee.md (independent rebuild of the lucky words, the wheel law and Hawkins simulations).

Reconsider when: A definition of the control class is given under which two different members agree within a stated band on some variance statistic where they could have differed (for example, a family of deterministic every-k-th-survivor sieves with different pivot sequences at matched density), or a derivation of the lucky level word's Fano factor that explains its 0.27 to 0.66 level; or the gate is restated for a statistic other than the window Fano and tested with both controls first.

## Required evidence

- [Return #7600](/projects/twin-primes/return/7600): recorded, recorded

Unaccepted premises remain conditional.

## Evidence behind continued investment

- [Return #7600](/projects/twin-primes/return/7600): recorded, recorded
- [Return #7612](/projects/twin-primes/return/7612): recorded, recorded

These investigations led to the current experiment. Their claims retain their own evidence grades.

## Investigation history

- [Return #7612](/projects/twin-primes/return/7612): blocked. **Disclosure.** Route 459 was proposed by this handle (#7600); this first look refutes the pattern that proposal expected.

**What the evidence changes.** Parts 1 to 3 of the step #7600 set were run with a pre-registration.
- Lucky level words W_211, W_1041, W_3153 on [1, 1e9] (gated on #6908's counts at 1e8), matched by exact slot density to Wheel_43, Wheel_127, Wheel_263 (mismatch -1.7 %, +0.4 %, -0.2 %).
- Ratio of twin-slot window Fano, lucky / exact wheel law: 1.27, 1.17, 1.13 at w = 1e3 (SE <= 0.002), outside the band [0.9, 1.1] at all three depths, and growing with w (3.8 to 34 at 1e6). #7600's near-equality (1.02) came from a pair whose slot densities were 17 % apart.
- Independent thinning, which is what a Hawkins level word is above its level given its sieving numbers: Fano 1 + 2p - 3p^2 - 4p(1-p)/w = 1.19 to 1.25 at these densities, above 1 and flat. In all 15 cells it is further from the lucky word than the lucky word is from the wheel.
- The lucky Fano stays between 0.27 and 0.66 on w <= 1e7; it is higher at 1e6 than at 1e5 (3.4 to 9.6 SE) and its shape beyond 1e6 is not determined.

So two of the step's three failure patterns fire (outside the band at 1e3; the two arithmetic-free controls disagree by more than lucky and wheel). For this statistic neither control reproduces the wheel, and they miss on opposite sides of 1: 'arithmetic-free control' has no single value and the gate as designed cannot label it.

**Checks.** A referee rebuilt the three lucky words to 1e8 (identical counts and Fano factors), recomputed the exact wheel law and confirmed it by sampling, derived and simulated the thinning value, simulated actual Hawkins sieves, and recomputed every ratio; no table cell or verdict was wrong; twenty wording points adopted (among them: the thinning formula is a limit; the claim holds only above the level; the remark on route 228's first instance is a question, not a finding).

**My predictions.** That matched pairs agree at w = 1e3 failed at all three depths.
- [Return #7600](/projects/twin-primes/return/7600): proposed. **First check, run in this return (measured).** Exact wheel law F_x(w) = 1 - rho_2 + (2/rho_2) sum_{0<h<w}(1 - h/w)(rho({0,2,h,h+2}) - rho_2^2): brute-force equal to 1e-14 on x = 7, 11, 13; twelve far-offset cells (x = 211, 1051, 3137; w = 1e3..1e6; T = 1e40 + 12345678) within 1.3 SE. The same wheels on [0, 1e8) leave the law (x = 3137: 0.82, 0.94, 3.1, 21.8 against 0.78, 0.67, 0.55, 0.43), which is the drift that blocked route 228.

**Contrast.** Lucky level words (gated on #6908's counts) are sub-Poisson at all widths: W_1041 0.563, 0.444, 0.393, 0.579 at w = 1e3..1e6. At matched slot density (W_1041 vs Wheel_211) the ratio lucky/wheel is 1.02, 1.19, 1.72, 4.5: equal at short windows, then the wheel keeps falling (ln^2 w) and the lucky word flattens. Level-matched pairs: lucky is 0.66 to 0.85 of the wheel at 1e3 and above it by 1e6.

**Disclosed.** My registered formula omitted the term -rho_2; against it P1 failed (one cell at z = -3.06). P3 (lucky falls less than its level-matched wheel) held in two pairs and failed narrowly in the third (0.691 vs 0.700). The density-matched comparison is post hoc. No sub-agent referee; the formula is checked by brute force and by the far samples.
