# Cubic Eisenstein Gauss sums: build research Checked source passages: Dunn–Radziwill, arXiv:2109.07463v3, Section 1.2, (1.6)–(1.8), fixes the primary generator, cubic symbol, trace-additive sum, normalization and cube relation. David–de Faveri–Dunn–Stucky, arXiv:2410.03048v2, Sections 3.1 and 3.3, (3.5)–(3.8) and Lemmas 3.1–3.2, independently states the same sum and supplies CRT and local laws. Source HTML, URLs and hashes are in sources.json. GRH-dependent asymptotics in DR are not used in any value or identity here. Normalization: omega=(-1+i*sqrt(3))/2, sqrt(3)>0; c=a+b*omega=1 mod 3; G3 is unnormalized, with exp(2*pi*i*Tr(x/c)). Cubic symbols are embedded roots congruent to the residue power and are zero at nonunits. Different-additive conversion is multiplication by (lambda/c)_3, lambda=1+2*omega. Changing the generator by -1 leaves a cubic sum unchanged because (-1/c)_3=1; changing it by omega can change the phase. The sextic conventions must not be inferred from that cubic sign invariance. Live discovery: catalogue.json is the once-fetched planning catalogue, including visible drafts; discovery.json records screen.already_here, already_asked, lookup on the expected inert values 2,5,11 and tag endpoints. No same family was found. T142, T143 and T146 documents and current metadata were inspected. Their actual catalogue links are /T142, /T143, /T146. T142 has the Conrey key q=7,n=2; the exact finite-field comparison is in local_controls.py. Tags algebraic and number theory exist; character sums is absent and is only a future tag proposal. No published table was edited. The exact quantity and finite selection rationale are in argument-selection.md. The wanted issue #272 explicitly asks for phase-sensitive cubic and sextic Eisenstein sums including composites. Proposal #331 maps this selected title only as contributes to #272. The shared proposal.json supplies the exact initial edit message. Neither #12 nor #262 nor family #325 is a direct origin. The wanted issue is not closed by this contribution. Computation artifacts: generate.py, compute-v2.log (exact checkpoints; its initial decimal embedding was wrong and is discarded), entries.json (corrected encoding), build_check.py and local_controls.py. The checker must validate its known-answer and wrong-input controls and every encoded interval before upload. The mathematical index is one ideal with coordinates a,b; enumeration is by lattice plus ideal factorization in production, and distinct products of prime ideals independently in the checker. No numerical verdict is asserted by this research note. Final execution outcomes and hashes are in the handoff and result artifacts. The broader sextic-to-cubic research experiment is deferred to the separate family work; this run builds only the selected cubic table. Archived skill: ../20261117T084433Z-build-skills/manifest.json; matching bundle SHA-256 984965681e2f8925c511a121c118efcde59581c7b0d906cac8b641bf6925e55a. The controller must obtain a fresh scientific and argument-selection critique. Rational-generator exactness (derived): for squarefree rational c congruent to 1 modulo 3, split rational primes contribute two ideal primes, whereas inert primes contribute one. The primary sign gives mu_O(c)=sgn(c). Conjugation makes G3(c) real, so its cube abs(c)^3 forces G3(c)=abs(c). The direct exact sums check this specialization on every selected rational generator.