<!-- All digits below are copied verbatim from the gate records of this work (2026-06-26);
     assembled for the blog 2026-06-30. -->

# LBL3E — sunrise-dressed box (3-loop light-by-light, 6 lines, all mass m)

Result status: all 10 masters function-level closed; the elliptic sectors are
anchored by two independent codepaths at 35.7–116.4 digits. One independent
oracle (the uncut sector-63 finite part) is still outstanding — see the last
section.

IBP reduction: 10 masters. Elliptic sectors {15, 55a–e, 63} (7 masters) share the
inherited Γ₁(6) sunrise curve, fibred over the box virtuality w (cusps
w ∈ {0,1,9,∞}, Kodaira fibres (1,2,3,6)). Polylog sectors {11, 51, 59}
(3 masters) are structurally polylogarithmic (genus-0 dlog; certified 7-letter
box alphabet × sunrise thresholds {w, w−m², w−9m²}).

## Closed form (elliptic sectors, m² = 1)

Every elliptic master is function-level closed as

    I_63^(k)(w) = psi1(w) * [ Eichler_f3(tau(w)) + c1^(k) ] + c2^(k) * psi2(w)

with psi1 = (2π/√3)·varpi0 and Eichler_f3 the Eichler integral of the weight-3
Eisenstein form f₃ on Γ₁(6), evaluated by cusp-regularized Eichler-integral
transport on the Γ₁(6) elliptic fibration from the anchor point w₀ = −2.

Homogeneous period: varpi0(w) is the holomorphic Frobenius solution of the
equal-mass-sunrise Picard–Fuchs operator (t ≡ w/m²):

    L_sun = t(t-1)(t-9) d²/dt² + (3t² - 20t + 9) d/dt + (t - 3),   varpi0(0) = 1.

Exact normalization (verified independently to 35.7–116.4 digits at 6 points):

    psi1(w)/π = (2/√3) · varpi0(w).

## Anchor data (w₀ = −2)

    psi1(w0)/pi      = 7.611662731988950350519990804725388336643423169012396509202301896585102328333879433893615374442008061260141404980209230074e-01
    Im psi2(w0)/pi   = 2.488419648503361355719549018767424794944767838784660283928203594493117871306060772810790410402587514677797794849488866035e-01

Eichler_f3(tau(w)) at the six Euclidean anchors:

    w = -1/4 : 1.38817844254639186702577855195321568326744350538663866688878666591556370279045032e-01
    w = -1/2 : 2.68186911702707030412935783456189579962722885432755027011623812250699845045685253e-01
    w = -1   : 5.05147907400042020563474169569555481648108762841180230822831544142570000713626975e-01
    w = -2   : 9.17475992194864665245709077614316962005880489156111841664359129102196314547952341e-01
    w = -3   : 1.27381749916280277395698391371279828563429643030477115073279931188086876905663904e+00
    w = -5   : 1.87857593822846368958413555588180240447466970720917871048665416215145295958234949e+00

## Boundary constants (cusp ring, conductor 3)

The leading boundary constant PSLQ-closes at 110.0 residual digits:

    B^(0) = 2·Cl₂(π/3) = 2.029883212819307250042405108549040571883378615060599584034978213553194952516488044272940708456513389891723655062719770803e+00

The sub-leading boundary constants B^(1..4) were determined numerically (120 digits) and are given by the
2F1 eps-derivative form of the sunrise boundary (the eps^k coefficients of B(eps) built from
2F1(-2eps,-eps;1-eps;e^{2 pi i/3}); lbl3e-master-evaluate.py gate G5). The PSLQ search in the conductor-3
cusp ring was run for B^(1) and B^(2) and found no relation for either (a partial ring hit);
B^(3), B^(4) were not searched. The per-master
(c₁, c₂) constants for the box-dressed sectors 55a–e/63 are fitted by a
variation-of-parameters constants fit whose mechanics were validated by a
synthetic round-trip to 116 digits.

## Validation gates (all PASS ≥ 30 digits)

| gate | min digits |
|---|---|
| homogeneous anchor cross-check (PF Frobenius × Γ₁(6) q-series, 6 pts) | 35.7 (92.8–116.4 at 5/6 pts) |
| leave-one-out PF-transport (6 anchors × 5 held-out) | 35.7 |
| elliptic-top leave-one-out (Eichler ΔI, anchor −2 → 5 held-out) | 35.1 |
| variation-of-parameters constants fit, synthetic round-trip | 116 |
| B^(0) PSLQ residual | 110.0 |

The 35.7-digit floor at w = −5 is q-series convergence at 500 terms
(|q| = 0.212), not a method floor.

## Remaining caveat

A from-scratch AMFlow oracle for the box-dressed sector-63 finite part is still
missing. The originally proposed maximal-cut gate was shown to be ill-posed
(2026-06-27): the D4 cut pins w = m² identically — a degenerate cusp fibre — so
a sector-63 maximal cut cannot see the elliptic modulus or the (c₁, c₂)
box-dressing constants either way. A fully independent confirmation of those
constants requires the full (uncut) sector-63 solve or sub-leading cuts; the
closed form above is otherwise gated as listed.
