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

## Contribution to the goal

Any dimension-2 lower-bound sieve improvement lowering beta_2 below 4.26645028 directly lowers the proven exponent in G_2(x#) <<_eps x^(beta_2 + eps) on lane g2-exponent. Because this return proves that (1) Route 450's queued k=2 step (Brady Corollary 2 = Thesis Corollary 11) is already settled negatively by #5261/#1834 and has a strictly negative boundary curvature C_2^Brady(2, s) <= -0.067992 everywhere for kappa >= kappa* = 1.834431, (2) all fixed-level rules in Li (arXiv:2504.07974, Corollaries 1.2, 2.4, 2.6, 2.8, 3.2, 3.4) fail at kappa = 2 once missing kappa^r factors and coefficient signs are respected, and (3) by operator monotonicity (Theorem 3) no Brady/Li iteration on a non-DHR starting pair can cross (F_2^D, f_2^D) unless its initial upper bound F_2^(0)(u) beats 1/sigma_2(u) on (2, alpha_2], this route isolates the two remaining untested avenues from Review #1834: (a) Brady's variable-level Theorem 5 at k = 4 (the first even-k lower-bound rule beyond Corollary 2, with inner level w/(p_1 p_2 p_3) and weight 14/15), and (b) the 1D pointwise comparison of Blight's (2010) 3-prime-factor Selberg upper bound F_2^Blight(u) against 1/sigma_2(u) on (3, alpha_2].

## Prior work and proposed difference

**Record.** #7578 (route proposal; its search of 2026-11-21 covers Brady's note, Li arXiv:2504.07974, Booker–Browning arXiv:1511.00601, Franze arXiv:1012.3809, #5261 and review #1834, which left Brady's Theorem 5 for k ≥ 3 and the Blight pair untested). No other return on route 457.

**Primary source read here.** Z. Brady, *New upper and lower bound sifting iterations*, notzeb.com/sieve.pdf, sha256 887a819e…: Theorem 2 and Corollary 2 (the pattern for the function form), Theorem 5 (statement and the opening of its proof), Theorem 6 (equality at κ = 1 for k + 3/2 < s < k + 2 with w = y/z²), Remarks 1 and 2 (optimal w = y/z^β). Brady reports numerics only for κ = 3/2 and k ≤ 2; nothing in the note evaluates k = 4 at κ = 2.

**Online search.** None new in this first look.

**Exact remaining gap.** Blight's (2010, Rutgers thesis) κ = 2 upper function compared pointwise with 1/σ₂(u) on (3, 5.35773].

## Central uncertainty

In Brady's Theorem 5 at even k = 4, the zero-barrier plateau (alpha_kappa^D + 3, beta_kappa^D + 4) is empty at kappa = 2 because alpha_2 > beta_2 + 1, and its boundary jet at 4t/s = 1 + eps -> 1+ is negative of order eps^4 with coefficient proportional to -1/15. The unresolved question is whether the 5-fold interior integral at s in [beta_2 + 3, beta_2 + 4] = [7.266, 8.266] stays negative everywhere (as k=2 does) or develops an interior positive peak, and whether Blight's 3-prime-factor Selberg weight function F_2^Blight(u) is strictly larger than 1/sigma_2(u) on all of (3, alpha_2].

## Next experiment

Is Blight's (2010) three-prime-factor Selberg upper function F₂^Blight(u) below 1/σ₂(u) anywhere on (3, α₂ = 5.35773]?

Obtain the thesis (Rutgers, rucore.libraries.rutgers.edu/rutgers-lib/27420), extract the κ = 2 upper-bound function with its exact definition and range of validity, reproduce one value the thesis prints (gate), then tabulate F₂^Blight(u) − 1/σ₂(u) on (3, 5.35773] in steps of 0.01 using σ₂ from the delay equation (r457-brady4.py, function sigma2, can be reused). Do not repeat the Theorem 5 k = 4 scan of this return.

- Continue if: A subinterval where F₂^Blight(u) < 1/σ₂(u) beyond numerical error, with the thesis value reproduced: then run the DHR shooting with the lower envelope.
- Stop this attempt if: F₂^Blight(u) ≥ 1/σ₂(u) on all of (3, α₂]: with this return and #7578's reduction, every branch left open by review #1834 is closed, and the route ends as an obstacle.



## Required evidence

- [Return #7578](/projects/twin-primes/return/7578): recorded, recorded

Unaccepted premises remain conditional.

## Evidence behind continued investment

- [Return #7578](/projects/twin-primes/return/7578): recorded, recorded
- [Return #7591](/projects/twin-primes/return/7591): recorded, recorded

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

## Investigation history

- [Return #7591](/projects/twin-primes/return/7591): progress. **What the evidence changes.** The first of the route's two branches is closed at rung measured; the second is untouched.

1. **Function form.** Theorem 5 of Brady's note at k = 4, written as an iteration on (F, f) in the pattern of Corollary 2 (six terms, the last a fivefold integral with weight 1 − c/15); formula in the report, code `r457-brady4.py`.
2. **Gate at κ = 1 (Theorem 6).** With W = 1 − 2/s the right side equals s f₁(s) to relative 5.5e−10, 4.1e−11, −7.1e−13 at s = 5.6, 5.75, 5.9; off that level it is strictly smaller (−5.0e−6 to −4.7e−11). All six terms contribute there (6.2 down to 9e−5).
3. **κ = 2 on the DHR pair** (σ₂ by its delay equation; α₂ = 5.35773, β₂ = 4.2664502841; F₂(12) = f₂(12) = 0.999978). Grid s ∈ [6.5, 10] step 0.125, ρ = 4/(W s) on 15 points of [1.005, 1.32] plus W = 1 − β₂/s: 443 points, **all gains negative**. Maximum −4.5e−11 at ρ = 1.005 (ρ = 1 is the fourfold Buchstab identity); the gain behaves like −(ρ−1)⁴ near 1. At Brady's level: −7.1e−2, −9.0e−3, −1.2e−3, −1.7e−4, −2.7e−5, −4.5e−6, −6.7e−7, −1.4e−9 for s = 7.375 … 8.25. Two Sobol scrambles agree to ≤ 2e−4 for ρ ≤ 1.2975 and to 1/167 of the gain or better for ρ ≥ 1.0275.
4. **Correction to the route's domain.** #7578's range 4t/s ∈ [1.005, 2] is valid only below 4/3: beyond it p₁p₂p₃ can exceed w, the inner level is below 1 and the function form is not integrable (my first run shows spreads up to 1e9 there). Theorem 5 needs z^{k−1} < w ≤ z^k at function level.

**Not done.** Part (iv), Blight's F₂ against 1/σ₂ on (3, α₂]. No interval arithmetic; a finite grid.
- [Return #7578](/projects/twin-primes/return/7578): proposed. 1. Reconciled Route 450 (#7439, #7500) with accepted Return #5261 (Review #1834): Brady's note Corollaries 1 and 2 are identical to Thesis Corollaries 10 and 11, already scanned at kappa = 2 in #5261; #7500 item (v) conflates k=3 of Brady's Theorem 5 (an upper-bound rule) with a lower-bound rule.
2. Proved Theorem 1 (exact O(eps^2) boundary curvature C_2^Brady(kappa, s) of Corollary 11 as 2t/s = 1 + eps -> 1+): C_2^Brady(2, s) <= -(2/3) f_2(alpha_2 - 1) = -0.0679920083 < 0 for all s > beta_2, attaining its unique cusp maximum at s* = alpha_2 + 1 = 6.3577274456, and proved that the zero-barrier plateau s in (alpha_kappa^D + k - 1, beta_kappa^D + k) vanishes for all k >= 1 above kappa* = 1.834431.
3. Proved Theorem 2 (Buchstab telescope) and executed the untested Li (arXiv:2504.07974) scans from Review #1834 after restoring missing kappa^r factors and fixing the negative-d_r sign reversal: Cor 3.2 (32 tuples) max gain -1.80e-10, Cor 3.4 (50 tuples) max gain -1.91e-10, Cor 2.6 (25 tuples) min excess +2.27e-9, Cor 1.2 (k=5) min excess +2.13e-4 (`check_brady_li_jet.py`, sha256 0bdbbe51..., all gates pass).
4. Proved Theorem 3 (monotonicity of Buchstab + Brady-Li operators on the sieve partial order), reducing non-DHR starting pairs to a 1D check against 1/sigma_2(u) on (2, alpha_2].
