Parity with Python

screamer.js is not a reimplementation of the Python screamer package. It compiles the same C++ operator kernels to WebAssembly and calls them through a thin JS binding layer, the same way the Python package calls them through a thin pybind11 layer. There is one implementation of every operator's numerics, shared by both languages.

That means:

• Same results. Every operator's output matches the Python package's output on the same input to 1e-9, verified by cross-language fixture tests.
• Same causal semantics. No operator looks ahead. A batch call and a step-by-step streaming call on the same data always agree, in both languages.
• Same warmup behavior. startPolicy options ("strict", "expanding", "zero", where an op exposes them), NaN warmup periods, and window-fill rules match exactly.

What differs between the two packages is the surface, not the math: JS uses camelCase op names and options (RollingMean, startPolicy) where Python uses snake_case (rolling_mean, start_policy), and the calling conventions follow each language's idioms (see Calling conventions for the JS side).

Where to go for the math

This guide covers the JS surface: how to call an op, how to build a pipeline, how to manage lifetime. It does not re-derive each operator's formula. For the full per-operator reference, including the underlying math, parameter semantics, and worked examples, see the Python documentation. Every op factory in the API reference here links back to it as well.