Changes in version 0.2.1 (2026-08-21)                  

CRAN resubmission

  - The Description field no longer opens with a redundant "an R", and
    single quotes are now limited to software and package names
    ('AstraeaDB', 'Apache Arrow Flight', 'arrow') rather than acronyms
    such as BFS or GQL. Acronyms that remain are spelled out on first
    use.
  - No \dontrun{} remains in the package. Two examples that need no
    server run unconditionally; the rest are \donttest{} guarded by
    astraea_server_available(), so they are a no-op without a server and
    genuinely execute with one.
  - Examples touching the optional Arrow Flight transport are wrapped in
    requireNamespace("arrow", quietly = TRUE), and 'arrow' stays in
    Suggests.

Bug fixes

  - UnifiedClient$query_df() no longer fails with "arguments imply
    differing number of rows" when a matched node lacks one of the
    requested properties, or when rows carry different property sets. An
    absent property now becomes NA and the result takes the union of
    columns across rows. This affected ordinary queries: MATCH
    (n:Person) RETURN n.name, n.age errored whenever any person had no
    age.

Documentation

  - Examples no longer reference hard-coded node and edge IDs such as
    1L. Each creates the nodes it needs and uses the returned IDs, so it
    is self-contained.
  - Examples that store embeddings size them from
    client$ping()$vector_dim instead of a fixed three-element vector,
    since a store pins its embedding width on first insert.

                        Changes in version 0.2.0                        

New features

  - Graph algorithms are now exposed as client methods, computed
    server-side: run_pagerank(), run_louvain(),
    run_connected_components(), run_degree_centrality(), and
    run_betweenness_centrality(). Each accepts an optional nodes
    argument to restrict the computation to a node subset.
  - Depth-first traversal: dfs() and the time-travel variant dfs_at().
  - Lookups: find_by_label(), find_edge_by_type(), and bulk
    delete_by_label().
  - Raw subgraph export via get_subgraph() and graph-wide graph_stats().
  - All of the above are available on both AstraeaClient and
    UnifiedClient.

Breaking changes

  - Removed the anomaly-detection methods (anomaly_check(),
    anomaly_stats(), and anomaly_alerts()); the AstraeaDB server no
    longer provides this operation.

Documentation

  - vector_search() results are now documented as returning distance
    (smaller is closer), with a legacy score alias retained for backward
    compatibility. semantic_neighbors() results likewise carry distance,
    while hybrid_search() returns a combined score.