Algorithm contributions

Numerical code must state its input domain, deterministic behavior, tolerances, and failure conditions. The current discovery path accepts finite numeric time series, estimates derivatives, builds scalar features, performs sparse regression, scores candidates, and renders symbolic expressions.

Add a focused unit test for invariants and an integration/scientific test for observable behavior. Fixtures must be generated from defined equations or recorded measurements with provenance; do not encode expected output from the implementation under test. Compare floating-point results using a stated tolerance and include conditioning or sampling assumptions.

Adding a method means wiring it through its configuration, validation, selection logic, diagnostics, documentation, and CLI/Python surface where appropriate. A method that cannot meet these requirements is not a supported option.