Tree Sets, Consensus, And Support

A tree set is a population of identified trees, not a loose collection of Newick strings. It may represent bootstrap replicates, posterior samples, gene trees, search candidates, or a deliberately assembled sensitivity set. The population meaning determines which consensus, support, distance, and uncertainty summaries are valid.

Preserve Four Denominators

Denominator	Question it answers
requested members	what population did the procedure intend to generate or inspect?
produced members	which runs, loci, states, or candidates emitted a tree?
admitted members	which produced trees passed taxon, root, burn-in, integrity, and eligibility rules?
clade-eligible members	which admitted trees could actually contribute to this clade, split, quartet, age, or branch summary?

These counts can differ legitimately, but none may disappear. A clade present in 80 of 100 admitted bootstrap trees has different support from a clade present in 80 of 80 successfully produced trees when 20 requested replicates failed. The summary must carry the population relevant to its numerator.

flowchart LR
    source["Tree-set source<br/>and sampling meaning"]
    inventory["Member inventory<br/>weights and failures"]
    admission["Taxon, root and<br/>burn-in/thinning policy"]
    summaries["Clades · splits · quartets<br/>distances · consensus"]
    selection["Selected tree or<br/>bounded conclusion"]
    artifacts["Tables, tree files<br/>and denominator record"]

    source --> inventory --> admission --> summaries --> selection --> artifacts

Identify The Population Before Summarizing

Population	Members mean	Required selection record
bootstrap trees	resampled-data inference replicates	requested, completed, failed and admitted replicates
posterior trees	retained states from a sampling process	chain identity, burn-in, thinning, retained indices and convergence context
gene trees	locus- or partition-specific histories	gene/partition identity, taxon mapping and failed loci
search candidates	evaluated topologies	search method, move set, starts, score and deduplication policy
sensitivity trees	plausible alternative structural assumptions	source of each alternative and eligibility rule

Combining these populations changes the scientific question. A posterior clade frequency is not a bootstrap proportion, and either can differ from the fraction of independently inferred gene trees containing that clade.

Use A Summary That Matches The Decision

Decision	Public bijux_phylogenetics.trees surface	Preserve with the result
count clades or splits	compute_clade_frequency_table, compute_tree_set_split_frequency_table	population, clade/split key, numerator, eligible denominator and policy
summarize compatible clades	compute_strict_consensus_tree, compute_consensus_tree, compute_majority_rule_extended_consensus	threshold, compatibility rule, accepted and rejected clades
select a representative posterior tree	compute_maximum_clade_credibility_tree	all candidate scores, tie rule and selected member identity
retain a credible clade set	compute_credible_clade_set	cumulative rule, included/excluded rows and ordering
compare tree-set members	compute_tree_distance_matrix, compute_posterior_tree_distance_diagnostics	distance definition, taxon/root policy, matrix labels and distribution rows
assess a reference tree	compute_reference_tree_clade_support, compute_reference_tree_quartet_support	reference identity, eligible observations and absent/incompatible cases
summarize posterior quantities	summarize_posterior_branch_lengths, summarize_posterior_node_ages	structural key, retained samples, missingness and interval definition

No summary is the tree set itself. Retain the admitted inventory so a reviewer can reconstruct its denominator and determine which information the summary discarded.

Consensus Is A Projection

A strict consensus retains only clades present throughout the admitted set. A majority consensus applies a frequency threshold. Majority-rule extended consensus additionally resolves compatible lower-frequency clades according to an ordering and compatibility policy. These outputs answer different questions even when they happen to render the same topology.

Record the method, threshold, tie ordering, root policy, taxon population, accepted clades, rejected conflicts, and any unresolved nodes. A consensus branch length or support label also needs a declared aggregation rule; it must not be inherited from whichever member happened to supply the topology.

Support Needs A Denominator And A Method

Support belongs to a structurally identified clade or split and a named replicate population. For every reported value retain:

• clade or split key and the admitted taxon population;
• support method and scale;
• number of eligible, completed, failed, and excluded members;
• numerator and denominator used for the value;
• absent, incompatible, or unresolved treatment;
• any weighting, burn-in, thinning, or deduplication policy.

Do not merge bootstrap support, posterior clade probability, quartet support, and gene-tree concordance into a single generic support field. Similar percentages can encode different experiments and uncertainty sources.

Separate Selection From Uncertainty

Maximum-clade-credibility selection, consensus, and quartet-puzzling produce a tree that can be consumed downstream. The selected tree does not erase the distribution from which it came. Comparative or ancestral analysis performed on that tree is conditional on the selection unless sensitivity across the tree population is evaluated separately.

flowchart TB
    population["Admitted tree population"]
    representative["Consensus or selected tree"]
    distribution["Frequencies, distances,<br/>modes and conflicts"]
    analysis["Downstream analysis"]
    claim["Qualified conclusion"]

    population --> representative --> analysis
    population --> distribution
    distribution --> claim
    analysis --> claim

Budget Without Hiding Members

Large tree sets require explicit processing budgets. The tree-set budget surfaces record selected members, limits, sampling or truncation policy, and the resulting processing population. A resource limit may justify a bounded subset; it cannot silently convert the subset into the full population.

Reject a summary when member parse failures disappear, taxon populations are mixed without a reconciliation policy, posterior selection is unspecified, or output rows cannot be traced to structural keys. A complete tree-set result includes both the requested population and the population actually summarized.