# Research: T733, Riemann zeta heat trace Checked sources are archived in sources/manifest.json (URLs and SHA-256), with exact downloads retained. Skill: ../20261117T060056Z-build-skills/manifest.json, matching bundle SHA-256 984965681e2f8925c511a121c118efcde59581c7b0d906cac8b641bf6925e55a. This is build evidence, not an independent verdict. - Connes, arXiv:2402.13082v1, (4)-(7), (13)-(15), (17), (25): Gaussian heat test, both pole terms, archimedean integral and all prime powers. Its operator interpretation assumes RH. No global RH assumption is used here. - Suzuki, On the Hilbert space derived from the Weil distribution, Section 3.1, (3.3): explicit formula over the full complex zero divisor, including multiplicities; gamma=complex frequency in that source. Its even Gaussian specialization gives the displayed prime/archimedean route. - Montgomery-Vaughan, Multiplicative Number Theory I, Section 12.2, Theorem 12.13, pp. 410-411: weighted integrability and bounded variation hypotheses cover the noncompact Gaussian. Put F(x)=2*pi*phi_alpha(2*pi*x), so its Phi(s) is exp(alpha*(s-1/2)^2/(4*pi)). This is a scale conversion derived here. - Hasanalizade-Shen-Wong, arXiv:2107.06506, Corollary 1.2, (1.5): for T>=e, |N(T)-T/(2*pi)*log(T/(2*pi*e))| <= .1038*log(T)+.2573*loglog(T)+9.3675. N counts all positive-height nontrivial zeros with multiplicity. Use the deliberately loose consequence N(u)<=u^2 for u>=300; the zero-tail integration is elementary and recorded in check_values.py. - FLINT zeta-zeros documentation: rigorous consecutive zero indices, isolation and zero-count completeness. Installed Sage's ComplexBallField.zeta_zeros(150) is probed in numerical_probe.py; index 138 is below 300 and 139 above 300. Current online docs are 3.7.0-dev; installed library version is recorded separately, not assumed equal. Normalization: xi(s)=s(s-1)*pi^(-s/2)*Gamma(s/2)*zeta(s)/2; sum exp(alpha*(rho-1/2)^2/(4*pi)) over all its zeros, both signs/multiplicities. alpha>0 real; stored alpha positive rational; one named quantity, one parameter. Fourier exp(-2*pi*i*x*y); phi_alpha=alpha^(-1/2)*exp(-pi*x^2/alpha); t=alpha/(4*pi). Absolute convergence follows from the critical strip and O(T log T) zero count; reality from conjugate zeros. This does not assert positivity for every width or an unconditional self-adjoint operator. Numerical design: prime/archimedean generator uses rigorous Arb quadrature with an entire hypergeometric quotient at x=0, whose limit is 1/4. After L, subtract atanh(exp(-L)) and bound the Gaussian remainder. Prime-tail inequality is independently derived by Lambda(n)/sqrt(n)<=log(n)/sqrt(n)<=1 (max 2/e), the decreasing integral test and completing a square. The erfc upper bound avoids cancellation in 1-erf(z). Independent check uses certified zeros and a symmetric unconditional tail (off-line zeros may have complex terms). Live discovery: catalogue.json (729 published/accessible drafts), discovery.json, lookup.json. already_here and already_asked were rerun; the only exact family proposal is #329. No heat-trace table found. T254 heat polynomials, prolate tables, Li coefficients and reciprocal moments have distinct definitions. T3 supplies certified ordinate controls; T31 and T36 are related transforms, neither computational dependencies. Current metadata for T3/T31/T36 was inspected; canonical links https://numberdb.org/T3, /T31, /T36. Tags L-function, special values, integral exist. Riemann hypothesis tag is absent; proposal only, no existing tables retagged. Approximate lookup samples .6759992, .0007057871, 2.490351e-7 returned zero numeric matches. Discovery script's first malformed lookup was repaired using the public search_real_ball wrapper; refusal retained. Provenance: live #275 inspected through GitHub; its explicit-functional checklist is the origin. #276 concerns heat deformation of xi and #324 Lehmer diagnostics, not origins of this table. proposal.json's initial_commit_messages was used verbatim at creation. Private claim T733 is in claim.json. No publication, offer, acceptance or old-table edits are authorized in this actor. Argument selection: argument-selection.md and selection-plan.json. All 46 reduced widths 1<=a<=b<=12 retained, intrinsic alpha=1 plus noncanonical reference sample of the other 45. No conclusions from sibling tables are reused. Proof-facing bounded experiment: preserve an exact archimedean remainder representation and test uniform-sign candidates on denominator<=8, then withheld denominators 9-12. Computed values and counterexamples are in remainder-results.json: both uniform signs are refuted in the training widths, and all 24 withheld widths were evaluated. The positive archimedean kernel alone is proved by coefficient comparison; it does not make the arithmetic remainder positive. No uniform full arithmetic theorem or RH claim is inferred from finite signs. Numerical checks completed: all 46 independent zero enclosures lie inside the stored decimal intervals; verify(sample=None) matched 46/46. Authenticated audit has no findings. See stored-check-results.json, verify-results.json and precision-summary.json. Source/error/coverage dependencies and exact hashes are recorded in evidence-manifest.json.