Release Notes

Version 0.2.5 - 2026-06-11

Added inference residual views to graphFigure for drawing current message residual magnitudes and reporting residual values in edge tooltips.
Added Gaussian marginal variance views to graphFigure for shading variable nodes by marginal covariance.
Added marginal value tooltip sections to inference-aware graphFigure output for Gaussian sum-product and discrete sum-product inference.
Documented inference-aware graphFigure and saveGraphFigure overloads for residual and variance visualization.
Added graph visualization documentation examples for inference diagnostics.

Fixed Gaussian min-sum frozen variable and edge message updates so frozen outgoing quadratic messages remain unchanged until explicitly unfrozen.
Fixed inference-aware discrete factor updates so invalid initializing unary updates are rejected before mutating the graph.

Added numerical correctness coverage for Gaussian schedules, damping, freeze/unfreeze message preservation, tree exactness and selected step-by-step sweeps, root refreshes, warm-start updates, dynamic stale-state rejection, residual partial updates, WLS agreement for correlated vector models, and discrete brute-force references.

Removed the redundant factor node ID line from graphFigure tooltips.
Replaced graph figure view depth with hops, using true bipartite graph hop expansion from selected variables and factors.
Replaced separate unary and multi-factor graph spacing layout options with scalar or per-gap rowSpacing and columnSpacing values.
Expanded scalar Gaussian factor covariances to isotropic covariance matrices matching the factor mean dimension.
Added messages!(...; schedule = :residual, updateFraction/updateCount) as the one-step counterpart to residual-scheduled gbp!.

Added a graph visualization example page covering figure layout, labels, focused views, styles, highlights, and tree views.
Clarified when Gaussian and discrete flooding and residual schedule objects are needed instead of the schedule keyword.
Tightened Gaussian graph docstrings for labels, component lookup, factor updates, and stale inference object guidance.
Tightened package docstrings for discrete graph construction, residual schedules, inference updates, WLS helpers, and graph visualization.

Added residual scheduling coverage for stale schedules, discrete min-sum tree overloads, and zero/negative iteration handling.

Added optional SVG hover tooltips to graphFigure output for variables, factors, and edges, with configurable summary and full detail levels.
Added depth-limited graph figure views for drawing selected variables and factors with nearby factors and variables.
Documented graph figure tooltip controls in the Gaussian and discrete factor graph guides.

Embedded generated graph SVGs with object so SVG hover tooltips remain available in generated documentation pages.

Fixed generated SVG graph figures so Documenter includes them in deployed HTML documentation.

Changed automatically assigned graph labels to compact forms such as x1 and f1 instead of x_1 and f_1.
Renamed isDampedEdge to areDampedEdges to reflect that the selector can match one or more edges.

Added rendered factor graph figures to the examples.
Added quick graph drawing sections to the Gaussian and discrete factor graph guides.
Documented damping selector rules for areDampedEdges, dampEdges!, and undampEdges!.

Added discrete finite-state factor graph construction and inference utilities.
Added iterative sum-product and min-sum belief propagation for discrete models.
Added tree-oriented factor graph views and exact forward-backward inference on tree-structured graphs.
Added flooding, sequential, and residual message scheduling utilities.
Added dynamic graph update support with stale-inference checks.
Added weighted least-squares validation, diagnostics, and printing helpers.

Reorganized the package source into graph, inference, schedule, and utility components.
Expanded Gaussian factor graph support for scalar and vector variables, linear Gaussian factors, and Gaussian belief propagation in moment, canonical, and min-sum form.
Reworked and expanded the documentation, including Gaussian models, discrete models, examples, schedules, validation, and API references.
Expanded the test suite across graph construction, inference, schedules, tree inference, validation, diagnostics, and printing.