Scientific Domains

The runtime is organized around scientific responsibilities rather than a single end-to-end pipeline. Tree and alignment contracts support likelihood, inference, comparative, ancestral, and parsimony methods. Simulation and datasets provide controlled inputs. Validation, benchmarks, and reporting make results inspectable.

flowchart TB
    inputs["Taxa · trees · alignments<br/>traits · partitions"]
    foundations["Topology · branch lengths<br/>state spaces · covariance"]
    likelihood["Likelihood, distance<br/>and Bayesian inference"]
    comparative["Comparative and ancestral methods"]
    histories["Ecology and phylogeography"]
    parsimony["Parsimony and topology search"]
    context["Simulation and datasets"]
    review["Validation · benchmarks<br/>reports · evidence"]

    inputs --> foundations
    foundations --> likelihood
    foundations --> comparative
    comparative --> histories
    foundations --> parsimony
    context --> likelihood
    context --> comparative
    context --> parsimony
    likelihood --> review
    comparative --> review
    histories --> review
    parsimony --> review

Read The Map As A Set Of Decisions

The same tree may appear in several domains, but each domain asks a different question and owns a different record:

Encounter with the tree	Decision	Record that answers it
intake	are topology, rooting, lengths, labels, and annotations admissible?	validated tree identity and reconciliation ledger
inference	which topology or parameter values optimize or sample the declared objective?	method result, search or chain state, and diagnostics
evolutionary analysis	what does the tree imply for trait covariance, ancestral quantities, or histories?	estimand-specific result with model and population
assurance	does the result satisfy invariants, recover known behavior, remain stable, or correspond to an independent reference?	check, benchmark, sensitivity, or evidence observation
reporting	which accepted result is projected for a consumer?	table or figure linked to its structured source and omissions

Moving an object between domains does not transfer authority. Intake can establish that a branch length is finite, but only the consuming method can say whether its units and assumptions are suitable. Inference can produce a numerically complete result, but only diagnostics and a declared claim can establish whether that result is usable evidence.

Track The Scientific Object That Changes

Most cross-domain errors come from carrying a value forward after the object that gives it meaning has changed. Use the object transition—not the command sequence—to decide when a new identity and review record are required.

Transition	Scientific meaning that changes	Record to carry forward
raw records → admitted taxa or characters	analysis population and denominator	included, excluded, aggregated, and refused records with reasons
source tree → pruned, rerooted, scaled, or transformed tree	branch/clade identity and covariance geometry	parent identity, operation, parameters, invariants, and new tree identity
alignment → trimmed or partitioned alignment	site population and model assignment	retained/removed sites, partition coordinates, alphabet, and new checksum
traits → transformed design matrix	estimand, intercept, contrasts, and coefficient meaning	formula, transforms, levels, column order, and row/taxon order
model configuration → fitted result	likelihood, prior, objective, and uncertainty convention	complete configuration, estimates/samples, diagnostics, warnings, and status
fitted result → table, tree, map, or figure	selection and visual encoding	source result identity, selected rows/marks, units, omissions, and transform

Two objects may contain the same visible labels or headline value while having different scientific identities. A pruned tree is not merely a shorter file; a standardized predictor is not merely the same column in another unit; a consensus tree is not one sampled tree with extra support labels.

Domain Map

Domain	Core scientific objects	Representative outputs
Trees, alignments, and taxa	topology, rootedness, branch lengths, support, sequences, partitions, identifiers	validated trees, reconciled alignments, topology and support ledgers
Tree operations and identity	rerooting, pruning, clade/branch identity, derivation and comparison	derived trees, population-change ledgers and invariant checks
Tree sets, consensus, and support	bootstrap, posterior, gene, candidate and sensitivity tree populations	inventories, consensus, representative trees, clade/split/quartet support
Likelihood and inference	state spaces, rate matrices, likelihoods, priors, search and sampling	fitted models, inferred trees, search traces, posterior samples
Distance-based inference	pairwise distances, matrix assumptions, joining and least-squares objectives	inferred trees, residuals, influence ledgers, bootstrap support
Bayesian inference	likelihoods, priors, proposals, chains, clocks and topology distributions	run manifests, checkpoints, posterior samples, diagnostics and predictive summaries
Comparative and ancestral methods	trait covariance, signal, regression, evolutionary modes, node states	coefficients, model comparisons, uncertainty and ancestral ledgers
Ecology and phylogeography	host, niche, region and coordinate histories on trees	transition, movement, outlier, exclusion and map ledgers
Parsimony	character costs, ancestral sets, topology moves, resampling	scores, reconstructions, consensus and support summaries
Search, model selection, and support	candidate populations, objectives, search state, selection rules and ties	traces, candidate ledgers, selected results and method-specific support
Discrete ancestral states and transitions	state order, Mk rates, root priors, node probabilities and branch opportunities	fit, node-state, transition, exclusion and tree-set stability ledgers
Simulation and datasets	generative models, known truths, reference and stress corpora	simulated trees/traits/alignments, dataset manifests
Simulation recovery and calibration	generator/estimator contracts, replicate states, regimes, metrics and independence	recovery rows, calibration summaries, failures and bounded conclusions
Validation, benchmarks, and reporting	invariants, tolerances, metrics, provenance, publication contracts	diagnostics, benchmark reports, manifests, tables and figures
Diagnostics and sensitivity	tree assumptions, temporal signal, alternative trees, models and policies	findings, stability rows, refusal reasons and decision records
Tree uncertainty and sensitivity	topology modes, unstable clades/taxa, run comparison and tree-population policies	conflict, instability, sensitivity and consequence records

Choose By Scientific Question

Question	Begin in	Continue into
Are the taxa, topology, lengths, states, and partitions coherent?	data and structure	the method that consumes the validated objects
What changes when a tree is transformed or a tree population is summarized?	tree operations	tree sets, consensus and support
Which topology or model parameters best explain observed characters?	inference	support, sensitivity, and assurance review
Was selection reviewable across every candidate, failure and tie?	search, model selection and support	method-specific diagnostics and benchmark evidence
How do traits vary with ancestry, predictors, or internal nodes?	evolutionary analysis	model comparison and tree/model sensitivity
Which discrete states or transitions remain supported across trees and models?	discrete ancestral states and transitions	tree uncertainty and sensitivity
Is an implementation recovering known truth or matching a reference?	scientific assurance	benchmark or Evidence Book artifacts
Is simulated recovery calibrated over the declared regimes and failures?	simulation recovery and calibration	validation and evidence promotion

Scientific questions commonly cross domains, but ownership remains local. A taxon mismatch is resolved before model fitting; a convergence failure is not recast as a reporting concern; and a visually persuasive figure cannot replace an uncertainty or sensitivity record.

Domain Handoffs

flowchart LR
    intake["Data and structure<br/>validated objects"]
    inference["Inference<br/>trees and parameters"]
    evolution["Evolutionary analysis<br/>traits and histories"]
    assurance["Scientific assurance<br/>checks and evidence"]

    intake --> inference
    intake --> evolution
    inference --> evolution
    inference --> assurance
    evolution --> assurance
    assurance -. failures and limits .-> intake
    assurance -. failures and limits .-> inference
    assurance -. failures and limits .-> evolution

Each handoff carries identities and assumptions, not only values. Tree scale, rooting, taxon order, state encoding, model parameterization, search or sampler configuration, and software provenance remain attached to downstream results.

Cross-Domain Handoff Record

Field	Why the next domain needs it
object identity	prevents a tree, matrix, alignment, or dataset revision from being substituted silently
taxon population and order	preserves joins, pruning decisions, matrix row meaning, and denominators
units and scale	distinguishes time, substitutions, distance, rates, transformed traits, and probabilities
model and state conventions	fixes alphabets, ambiguity, partitions, covariance, transition rules, and parameterization
execution state	distinguishes complete, partial, refused, failed, and externally unavailable work
diagnostics and exclusions	prevents downstream presentation from retaining only successful values
provenance	identifies implementation or engine owner, version, configuration, and source inputs

A downstream method may add context, but it must not erase upstream identity or reinterpret an unresolved condition as an ordinary value.

Recognize When A Handoff Changes The Question

Some transformations produce a new scientific object or estimand rather than merely preparing the old one. The downstream record must name that change so a result is not attributed to the original question unchanged.

Handoff decision	Original question	New question that must be declared
prune to shared taxa	relationship on the declared taxa	relationship on the retained intersection
trim sites or exclude sequences	inference from the admitted character matrix	inference from a selected matrix under a removal policy
root, reroot, or rescale a tree	topology or process under the supplied tree identity	topology, direction, or rate under a transformed tree
standardize, log, aggregate, or impute traits	effect on the observed response and population	effect on a transformed or reconstructed quantity
select a model after screening candidates	fit of one predeclared model	selection-conditional result over a candidate set
summarize across trees, chains, or replicates	result for one execution state	distribution or consensus over a declared population
normalize outputs across implementations	engine- or method-specific quantity	correspondence under an explicit mapping rule

The transformation can be scientifically valid. The defect is allowing the new population, scale, conditioning event, or estimand to inherit the old label without a parent-child record and an interpretation update.

Attribute Failure To The Deciding Domain

Parsing and identity defects belong to data and structure. Undefined likelihoods, search exhaustion, or poor chain behavior belong to inference. Model sensitivity and unidentified covariance or transition assumptions belong to evolutionary analysis. Missing independent truth, reference incompatibility, or incomplete denominators belong to scientific assurance. Reports carry these states; they do not become their owner.

Leave With The Owning Record

After the domain decision	Continue with	Preserve
execute an owned or external method	the product handbook	computational owner, input identity, configuration, status, diagnostics, artifacts
compare implementations	the verification matrix	both owners, selected cases, normalization, observations, tolerance, denominator
evaluate a study statement	the Evidence Book	claim, source, primary outputs, checks, freshness, verdict, limitations
publish a table, figure, or package	publication artifact review	structured source, visual encoding, inventory, consumer acceptance, revalidation identity

Domain coverage states what the runtime can express; it is not an assurance shortcut. Native methods and external engines retain different computational owners, and neither acquires study evidence until a governed claim consumes the exact result.