NEWS

treehunt 0.0.2 (2026-10-16)

Document return classes, output components, and their interpretation for the eight functions identified during CRAN review.
Require an explicit output_dir in run_ga_batch(). Calls that previously relied on the default "results" directory must now supply that argument.
Replace both disabled genetic-algorithm examples with small, self-contained examples that run during package checks and show the actual returned fields.

Return self-normalized estimates of the uniform combination score distribution, with a separate reference conditional on support strictly greater than beta.
Return the weight-based effective sample size for both references. Weights use exact capped scores before histogram binning and are rescaled for stability.
Add weight_ess() and uniform_score_reference() for explicit post-processing.
Add optional burn_in and store_trace arguments to the generic and PWP MCMC interfaces. Rejected moves contribute repeated states to the retained sample.

Correct unsigned subtraction in the hypergeometric upper tail: zero observed exposed events now give score zero, instead of an erroneous infinite score.
Both MCMC proposal types now operate on fixed-size subsets of distinct nodes, including ancestor-descendant pairs. Global proposals are uniform; local proposals exclude occupied destinations and use the corresponding Hastings correction. These corrections change trajectories from earlier development versions even when the same seed is supplied.
A positive prob_type1 is required to connect the observed state space. Initialization constructs a supported state and reports impossible sizes.
Correct hierarchical membership when a single observed descendant covers multiple selected ancestors. Genetic-algorithm validity rules are unchanged.
Exhaustive size-two references enumerate supported unordered pairs once, excluding self-pairs, and use unit weights. Previous reference outputs should be regenerated before comparison with corrected MCMC output.
Sort within-patient observations before computing medians in Wilcoxon post-processing and residual scores; scores no longer depend on row order.
Expose optional id_column in the depth-vector MCMC interface for continuous patient-level scores.
Validate finite temperatures, score caps, integer size controls and burn-in; support n_results = 0 and allow user interruption during long runs.

Align the package declaration with the maintainer-selected GPL-3 license.
Require C++17, which covers the language features used by the package.
Make saved-result helpers independent of implicitly attached tidyverse packages.
Guard traversal above tree roots and empty dissimilarity matrices; compute parent links for depth-vector trees as for data-frame trees.
Declare optional visualization dependencies and report missing dependencies.
Replace placeholder help pages; document the capped target and reference space.
Use explicit exports, preserving documented public helpers and keeping native engines and test helpers internal.
Replace the test dependency on emcAdr with an independent synthetic tree.
Add tests for weighting, ESS, filtering, burn-in, proposal support and exhaustive calibration, and a multi-platform R CMD check workflow.