Deterministic by Design
Cryptographically Verifiable
Geometric Anomaly Detection
All 18 Modalities

One Sheffer kernel. 18 modalities. Every benchmark, exactly as measured.

Six compete head-to-head with stochastic AI. Twelve more are kernel-native capabilities with no mainstream machine learning equivalent. Precise figures, no rounding up.

The 6 -- compete with stochastic AI

Head-to-head benchmarks against established stochastic baselines.

ModalityKey resultvs.
Neural Networks97.78% MNIST accuracyReLU 96.89% / GELU 97.03% / Swish 96.69%
Symbolic Regression43W / 31T / 26L on 100 AI Feynman equationsPySR 1.5.10, matched wall-clock
Symbolic Regression (high-dim)22W / 0L at 5+ variables -- sweeps every casePySR; Sheffer mean R2 0.92+, PySR 0.67-0.84
Reinforcement Learning7/8 episodes stabilized, mean return -16.50LQR: 3/8, -24.71 -- 21-parameter policy
Neuro-Symbolic ExecutionHeating R2 0.9963 / Cooling R2 0.9944Ridge: 0.9335 / 0.9313
Neuro-Symbolic ExecutionConcrete R2 0.8777Ridge: 0.4342
Generative Models5/5 UCI datasets beat RealNVPCanonical MAF-split benchmark
Generative Models72,102-79,051 samples/sec at batch 1024Quadro P1000 (two-gen-old GPU)
Generative Models1,064/1,064 bit-exact Python/RustZero drift across 96 coupling layers
Bayesian Optimization6W / 7L full suite; 4W / 0L held-outscikit-optimize gp_minimize
Bayesian Optimization11.5x lower regret on constrained Rosenbrockscikit-optimize; closed-form Constrained EI
Bayesian Optimization5-20x faster across all problemsCPU only

The 12 -- kernel-native, no mainstream equivalent

Capabilities enabled by the Sheffer kernel's monotone closed form, algebraic closure, and commitment discipline -- not portable to ReLU, GELU, or attention stacks.

ModalityKey resultvs.
Compositional ProgrammingResidual ~1e-7 at corners (AND/OR/XOR/NAND/NOR)float32 noise floorDeep-dive
Compositional Programming -- addMax delta ~7.5e-7 on 1,000 probesfloat32 ULP limitDeep-dive
Verifiable Inference0.0 max delta Python/Rust, 10,000 inputsno stochastic equivalentDeep-dive
Verifiable Inference128-byte proof, 2.7ms verifyno stochastic equivalentDeep-dive
Compositional Dynamical InferenceNRMSE 0.048 at N=64 -- beats 0.05 targetNeural ODE: never below 0.05 within N=512Deep-dive
Compositional Dynamical InferenceLorenz-63 explained variance >0.9999single observed coordinateDeep-dive
Rigorous Enclosure Engine0 violations in 1e7 probessampled confidence bands
Exact Expectation Engine2056-3349x speedup vs 1e9-point MCMonte Carlo UQ
Exact Expectation EngineExact Sobol agrees with SALib N=8192seed-dependent MC sensitivity
Behavioral Equivalence200k rewrite variants, 0 mismatchesbyte hash != function identity
Exact InversionRound-trip worst 3.04e-10 on 1e6 probesReLU preimage NP-hard
Distribution FingerprintsPower 0.367 vs RBF-MMD 0.316stochastic MMD baselines
Conformal PredictionCoverage within +/- 1% at three levelsposterior sampling calibration
Deterministic Spectral EngineECG AFib pooled test AUC 0.8637stochastic spectral pipelines
Atum Tensor14/16 cells at R2 >= 0.99stochastic tensor regression
ASM (Algebraic Shadow Models)98.21% label retention at 40.11% paramsteacher model footprint

See it applied to financial model risk

The Financial MRM use case shows how these benchmarks translate into an SR 11-7-ready deployment.

View Financial MRM Use Case