The Silver Web: A Working Pressure–Möbius–Silver Diagnostic

Abstract

The Silver Web is a working Pressure–Möbius–Silver diagnostic. It answers the practical question a single score cannot: which part of a finite binary model is wrong, and what measurement should come next? two invariant response anchors → Pressure calibration; sixteen-context response table → Möbius support and selected action; anchor tomography → Silver spectrum test; three typed streams → one complete fault-set diagnosis. The direct three-stream result is the primary proof of operation. Across all eight none/P/M/S fault combinations, the source-frozen base run identifies 11,505 of 11,520 complete fault sets. It confirms 5,758 of 5,758 support-selected pre-generated actions with zero false M confirmations and safely stops all 960 declared out-of-domain controls. A separately generated five-vertex structural run identifies 5,758 of 5,760 fault sets and confirms all 2,880 selected actions. An exact raw-input collision theorem proves that each stream is necessary for its designated label and that the full three-stream map is injective on the declared diagnostic domain. Silver is mathematically independent of entropy and capacity. For the interior spectrum p0=(1/2,1/4,3/20,1/10), an exact rank-three determinant is nonzero while the entropy-capacity Jacobian has rank two. The implicit-function theorem therefore gives an exact smooth curve on which entropy and capacity are identical but the Silver higher-jet residual changes; no single-valued function of entropy and capacity can reproduce it, even locally. The deterministic exact-arithmetic certificate passes 22/22 gates. A 157-spectrum fixed-(S_e,VarK) construction with residuals 0.29384–3.0048 supplies global numerical corroboration, while an equal-local-spectrum witness has carrier orders 1 and 2. The joined fingerprint changes decisions on physically distinct states. An 18-state seed-0 battery contains 18/18 unique physical states and gives 14/18→17/18. A distinct historical-seed-3 24-state battery gives 15/24 capacity-only, 16/24 with Silver, 18/24 with carrier, and 19/24 with all three, removing three capacity-only confusion types and introducing none. Both batteries are invariant across 200 record orders. Across every seed 0–99, the six-family panel has 1,800/1,800 unique rows and the eight-family panel has 2,400/2,400, with mean capacity-to-full gains of 3.31/18 and 2.89/24; the six-family rows are the corresponding subset of the eight-family generator, not additional independent states. On 160 unique training states and 320 disjoint held-out states, all-minimum-majority scoring gives 226/320, 228/320, 277/320, and 277/320 for the same four bundles, invariant across 200 training orders with no ambiguous class vote; the one- and two-cut SCP atlas reaches 320/320. The exact common response is generated by modular pressure. Φ_p(t)=ln Tr(ρ^(1−t)) has derivatives Φ_p^(m)(0)=κ_m(K). Its binary carrier is the one-coordinate curve ν(g)=(1+e^g)^−1 with g=ln(2)/z, while Q=d², v=d, L=E+2C, Möbius path/cycle atoms, and product-root addition carry the same coordinate through the finite bridge. The declared algebraic continuation of the cycle formula at length one equals Φ's binary coordinate Π_z; ordinary simple-cycle graph checks begin at length three. The validator recovers the dictionary on 16 inverse rows to maximum residual 4.5427×10^−15, reproduces six Monte Carlo response rows to maximum relative Var(L) error 0.0280838, and passes 82 formula/derivative checks. The twelve-record 9/12→10/12→11/12→12/12 chamber remains as a mutation-guarded integration fixture after these direct, exact, and empirical results. Rights are file-scoped: prose, figures, and reported results are CC BY 4.0; executable source code is Apache-2.0, as mapped in the packaged LICENSE.