The pure-Python mathematical core

omnibias-core owns backend-independent activation specifications, derivative coefficients, jet algebra, numerical support, and scoped proof and interval infrastructure.

The core does not require a tensor framework. Backend packages instantiate concrete specifications and tensor operations. A shared recurrence is a mathematical common source; it does not itself prove identical floating-point results for every backend operation.

Start with the coefficient example, then inspect the core API. Choose a focused module from that index rather than importing an implementation-private path.