theorem MathlibNt.SieveTheory.LiLiuOnePlusOneNine.GoldbachBig.goldbachWeight_paperSplit_authorG11_consumed_small_epsilon (δ : ℝ) (hδ : 0 < δ) : ∃ (ε₀ : ℝ), 0 < ε₀ ∧ ε₀ ≤ 2 / 15 ∧ ∀ (ε : ℝ), 0 < ε → ε < ε₀ → ∃ (N₀ : ℕ), 4 ≤ N₀ ∧ ∀ (N : ℕ), N₀ ≤ N → Even N → ↑(goldbachWeightG6 (goldbachDifferenceCarrier N ε) N (↑N ^ (4 / 53)) (↑N ^ (4 / 33)) + goldbachWeightG7 (goldbachDifferenceCarrier N ε) N (↑N ^ (4 / 53)) (↑N ^ (4 / 33)) (↑N ^ (3 / 11)) - goldbachWeightG12 (goldbachDifferenceCarrier N ε) N (↑N ^ (4 / 53)) (↑N ^ (4 / 33)) (↑N ^ (3 / 11))) + (124341093 / 200000000 - goldbachB9PaperSplitIntegral - 10191 / 100000 - δ) * (SingularSeries.liuSingularSeries N * ↑N / Real.log ↑N ^ 2) ≤ 4 * ↑(D19 N)

The already proved sharper S3, S4 and B10 constants are retained without recomputation. The full paper-split S5 bound and the certified actual G11 bound are consumed in the same original weight inequality. The author bound consumes its own third of the error budget; no constant is added back to an already weakened ledger. Only the literal G6 + G7 - G12 remains. This is a signed counting bound, not an assertion of final positivity.