NEWS.md
In hilldiv3: Integral Analysis of Diversity Based on Hill Numbers

hilldiv3 (development version)

Breaking: the out = "matrix" form of the per-sample functions (hilldiv(), hillprof(), hilleven()) now returns samples in rows and diversity orders (q0, q1, ...) in columns, instead of the previous orders-in-rows/samples-in-columns layout. This matches the usual observations-in-rows convention and joins cleanly with per-sample metadata. Code that indexed these matrices by ["q0", ] should switch to [, "q0"] (or transpose). The component/metric matrices from hillpart(), hilldiss() and hillred() are unchanged.
hilldiv() and friends now accept count tables whose first column holds taxa names: a leading non-numeric column is automatically promoted to row names (instead of failing with "Count data must be numeric"). Any other non-numeric column raises a clear error asking the user to fix the input.
Verbose output now reports the number of taxa and samples being analysed.

Complete redesign of the package built on a tested, isolated compute engine.

Math separated into an internal, unit-tested engine (hill_alpha(), hill_partition(), hill_beta_to_dissim()); the user-facing hill* functions are thin wrappers around it.
Single validation/alignment layer (prep_data()) that checks names with setequal() and reorders data to match the tree/distance matrix, fixing the silent-misalignment behaviour of hilldiv2.
New as_hill_input() adapter accepting matrices, data frames, tibbles, phyloseq and TreeSummarizedExperiment objects.
Fast ape-based post-order tree traversal replaces geiger::tips().

hillprof() — diversity profiles across a sweep of q values.
hilleven() — evenness from Hill numbers.
Real match_data() helper (previously only referenced in the docs).
Tidy output by default. hilldiv(), hillpart(), hilldiss(), hillsim(), hilleven(), hillprof() and hillred() now return long-format data.frames with print(), plot() and (when ggplot2 is installed) autoplot() methods. Pass out = "matrix" for the legacy shape.
Explicit type argument — type = c("auto", "neutral", "phylogenetic", "functional") on every entry point; "auto" keeps input-based detection, an explicit value asserts and validates the diversity type.
hillpair() computes the type-specific structure once over all samples and reuses it per pair (no more full re-partition per pair), and reports a progressr bar when available.
Bundled, documented example data: gut_counts, gut_tree, gut_traits.
reference argument on hilldiv() — reference = c("pool", "sample") selects the reference tree depth for phylogenetic Hill numbers. "pool" (default) reads every sample at one common depth T = mean(T_j) so values are mutually comparable (hilldiv2's multi behaviour); "sample" reads each sample at its own depth T_j. The two coincide on ultrametric trees. The option is deliberately absent from hillpart(), where T is fixed at the Chiu et al. (2014) mean per-sample depth — the unique value for which gamma / alpha is a valid decomposition with beta in [1, N].

Phylogenetic hilldiv() alpha corrected. The per-sample phylogenetic Hill number previously raised branch length L_i to the power q along with abundance ((L_i a_i / T)^q), which matched no standard quantity and disagreed with the partition engine. Branch length is now a linear weight ((L_i / T) a_i^q, Chao et al. 2010), so q = 0 recovers Faith's PD / T and the per-sample value matches hillpart() alpha at a single sample. This changes phylogenetic hilldiv() and hillprof() outputs relative to earlier 3.0.0 development snapshots.

Added test-coverage (Codecov) and lint (lintr) GitHub Actions workflows.
Added golden-value tests cross-checking the engine against vegan and hand-computed constants, plus edge cases (single taxon, empty sample, q = 1).

The hilldiv(), hillpart(), hilldiss(), hillsim(), hillpair(), hillred(), tss() and traits2dist() names from hilldiv2 are preserved. The default return shape is now a tidy data.frame; use out = "matrix" for the hilldiv2-style matrix.