Golden values

verification/data/abcloud_golden.json freezes 206 reference constants in 37 groups — the exact numbers every implementation of the 10-language verification stack (Python, Julia, C++, Rust, Go, Haskell, Fortran, JavaScript, R, MATLAB) must reproduce. All of them are searchable in the dashboard's Golden values tab.

Selected groups

Group	What it freezes	Size
D3	Dyson–Mehta Δ₃(L) at L = 5, 10, 20, 50	4
K_zeros	form-factor checkpoints K(t), t = 0.2…3.0 + RMS envelope	6
S2	number variance Σ²(L), L = 5…100	5
ad_zeros	Byers–Yang flux defect A₂ = 61.4924…	1
bN	the b(N) ladder, N = 100…10000 (+offset variant)	7
byers_yang	max defect on the L = 8 loop	1
gram_point	Gram points g₁…g₁₀₀₀₀	8
gue2_cdf / gue2_pdf	exact GUE CDF/PDF grids, s = 0.05…5.0	10 + 10
goe_cdf	GOE CDF grid	10
ham_*	first eigenvalues of three frozen Hamiltonians (clean / open / 2-vortex)	3 × 7
mon_entropy	entanglement entropies over 36 (L, α, c, f) configurations	36
rh_metrics	⟨r⟩, Δ₃, p, n_cb for two lattice runs	12
topo_tee	topological TEER observables	18
ks_zeros	Montgomery KS D and p vs the zeros	2
meta	provenance: γ₁, γ_N, Julia version, zeros file	5

How they are used

• every implementation loads the same JSON and compares with its own output under the tolerance protocol of verification/;
• the CI runs the quick protocol on every push;
• the dashboard's Golden values tab reads the same file (exported as golden_values.json) and offers search over group, key and value.

Regenerating

The golden file is produced by the Julia suite itself (code/ab_cloud_full.jl); it is frozen with the run manifests — it never regenerates silently. To re-export the dashboard copy after a new suite run: python3 scripts/prepare-dashboard-data.py.