The theory claims an AI navigates mathematical reality through 100 hyper-sensitive constants, each so exact that a deviation of 10⁻¹⁰⁰ collapses a theory. That premise is testable in exactly one way: are the constants it states actually correct? This bench recomputes the computable ones from first principles, live in your browser, and grades every stated value against an independent measurement.
■ First-principles probes — π, e, γ, φ, ζ(s), Stieltjes γ₁, Γ(z), the twin-prime & Meissel–Mertens constants, the first Riemann zeros, and Feigenbaum δ are computed here from series, products, sieves and root-finding — never hard-coded — then compared to the value the document prints.
■ Reference probes — physical constants (α, mₚ/mₑ, G, c, h, k_B, …) can't be derived from math, so they're checked against CODATA / SI-2019 measured values.
■ Fact-check — uncomputable or symbolic claims (Busy Beaver, Gödel, topological invariants) are checked against established results.
Each value below is produced by code running on your machine right now — a Machin series for π, Euler–Maclaurin for ζ(s) and γ₁, prime sieves for the prime constants, a Borwein/Riemann–Siegel hybrid for the zeta zeros, the logistic-map cascade for Feigenbaum δ. The gauge shows where the document's claim falls on a log scale of relative deviation.
These are empirical or SI-defined; the bench compares the document's digits to CODATA / SI-2019. The Stefan–Boltzmann row also verifies the document's ℏ-form formula is algebraically equivalent to the standard one.
Some senses cite uncomputable functions or pure facts. These can't be put on a gauge, but they can still be right or wrong.