Host Association And Niche Evolution

Host-switch and niche-transition reviews transform discrete tip observations and ancestral-state probabilities into node, branch, count, rate, clade, constraint, and exclusion records. They describe histories conditional on a tree and state model; they do not directly observe transmission, adaptation, or ecological mechanism.

flowchart LR
    observed["Observed host or<br/>niche states"]
    model["ER · SYM · ARD<br/>root · constraints"]
    nodes["Node probabilities<br/>ambiguity"]
    branches["Branch changes<br/>certainty"]
    summaries["Counts · rates<br/>clades · unsupported claims"]
    interpretation["Bounded ecological<br/>interpretation"]

    observed --> model --> nodes --> branches --> summaries --> interpretation

Define The Observation Unit Before The State

A tip label such as host = bat or niche = marine may summarize very different evidence. The analysis contract must identify what was observed and how it became one state per admitted tip.

Source condition	Required evidence	Why it changes interpretation
one observation per sampled individual	specimen identity, collection context, taxon mapping	the state is directly tied to one sampling unit
several observations collapsed to one taxon	aggregation rule, counts, conflicts, uncertainty	a taxon consensus can hide within-taxon variation
several simultaneous host associations	multi-state or ambiguity encoding, not an arbitrary preferred host	co-occurrence is different from missingness or uncertainty
niche derived from environmental measurements	variables, transformation, thresholds, and provenance	discretization creates the state vocabulary being reconstructed
literature- or database-derived association	source record, evidence class, reconciliation, access identity	reported association strength and sampling coverage may differ

The admitted population is therefore the set of reconciled observation units, not merely the set of tree labels. Every exclusion, aggregation, and state coercion changes the branches on which a switch or shift could be inferred.

Select The Owning Contract

Question	Public surface	Distinct retained evidence
Where are host changes reconstructed?	summarize_host_switching	host-state nodes, branch switches, counts, constrained/unconstrained fits, unsupported claims
Where and how often do niches change?	summarize_niche_transitions	niche-state nodes, transition rates, branch changes, counts, clade shift burden

Both surfaces preserve exclusions and warnings. Neither converts a most-likely state change into a certain historical event.

State Vocabulary Is Part Of The Model

Retain raw and normalized states, state order, missing/ambiguous coding, allowed-transition policy, ER/SYM/ARD parameterization, root treatment, and taxon reconciliation. Merging rare states or changing an “unknown” code alters the rate matrix and every downstream transition opportunity.

Host constraints are especially consequential. Compare constrained and unconstrained fit records and preserve UnsupportedHostSwitchClaimRow entries. A forbidden transition is not repaired by omitting the branch from the headline count.

Read Probability Before Change

A branch change joins two uncertain node reconstructions. Review parent and child probability vectors, their overlapping plausible state sets, confidence, certainty class, branch identity, and support before treating the row as a switch or shift.

Record	Question answered	Not established
node row	which states are supported at one node?	an observed ancestral ecology
branch row	how do parent and child reconstructions differ?	a witnessed event or mechanism
count row	how many retained rows fall into each transition class?	rate without an opportunity or time denominator
rate row	what transition intensity follows from the declared model?	empirical frequency independent of model
clade row	where do retained shift burdens concentrate?	causation or comparable opportunity across clades

Preserve Opportunity And Repetition

Raw counts depend on branch number, branch length, state availability, missing taxa, and certainty policy. Compare clades only with their descendant and internal-node populations, changed-branch count, transition diversity, and repeated-shift definition visible.

A dominant transition can arise from one strongly supported change or several uncertain changes. Preserve certain, uncertain, and strongly supported counts separately.

Separate Association From Causation

Coincidence between a reconstructed host switch and diversification, geographic movement, phenotype, or outbreak does not establish a causal path. Such a claim requires an explicit joined design, temporal and sampling assumptions, alternative explanations, and sensitivity to tree and state uncertainty.

Appropriate language is “the declared model reconstructs a host change on this branch with this certainty.” Avoid “the lineage adapted to the host here” unless independent evidence supports adaptation and timing.

Publication Packet

Retain the tree, observed-state table, state vocabulary, root and transition model, constraint identity, node probabilities, rate rows, branch changes, counts, clade summaries, constrained/unconstrained fits, unsupported claims, exclusions, warnings, and artifact inventory.

Refuse a clean ecological conclusion when states cannot be reconciled, constraints are missing or contradictory, root/model sensitivity reverses the result, node ambiguity makes the branch classification unstable, or excluded taxa change the relevant opportunity population.