Protocol Capability Matrix

This page records a smoke-test-backed snapshot of selected public PyRecEst classes and the protocol-style capabilities they currently provide.

The matrix is intentionally conservative:

• it checks a small representative subset of public classes;
• it records current behavior, not planned behavior;
• it treats methods that are present but raise NotImplementedError or an unsupported-operation ValueError as unsupported for that capability;
• it does not prove mathematical correctness, numerical accuracy, or backend parity.

The corresponding tests live in tests/protocols/ and should be updated when a class intentionally gains or loses one of these capabilities.

Distribution capabilities

Class	dim	input_dim	pdf	ln_pdf	sample	mean	covariance	convert_to	approximate_as	from_distribution
GaussianDistribution	yes	yes	yes	yes	yes	yes	yes	yes	yes	yes
LinearDiracDistribution	yes	yes	no	no	yes	yes	yes	yes	yes	yes
VonMisesDistribution	yes	yes	yes	yes	yes	yes	no	yes	yes	no
CircularUniformDistribution	yes	yes	yes	yes	yes	no	no	yes	yes	no

Notes:

• LinearDiracDistribution has a pdf method through the current class hierarchy, but the method is intentionally undefined for a continuous density interpretation and raises NotImplementedError.
• CircularUniformDistribution.mean() is not considered supported because the circular uniform distribution has no unique mean direction.
• convert_to and approximate_as are checked as structural conversion-entry capabilities. The matrix does not claim that every target representation is available for every source distribution.

Filter capabilities

Class	dim	filter_state	get_point_estimate	predict_linear	update_linear	predict_nonlinear	update_nonlinear	predict_model	update_model	update_nonlinear_using_likelihood	history recording
KalmanFilter	yes	yes	yes	yes	yes	no	no	yes	yes	no	yes
UnscentedKalmanFilter	yes	yes	yes	yes	yes	yes	yes	yes	yes	no	yes
CircularParticleFilter	yes	yes	yes	no	no	yes	no	yes	yes	yes	yes

Notes:

• Filter capabilities are checked at class level to avoid running prediction or update algorithms in this smoke-test layer.
• history recording means that both record_filter_state and record_point_estimate are available.
• Backend-specific algorithm support is documented separately in the backend compatibility guide; this matrix only records the public entry points exposed by the classes.

Maintenance rule

When a new public protocol module lands, keep this matrix aligned with it. For example, when distribution-specific protocols such as SupportsPdf or SupportsSampling are introduced, this page and the tests should use the same capability names and semantics.