Skip to content

Artifact Contracts

The runtime writes files that are meant to be read by people, automation, or both. This page explains which output families are safe to depend on as public artifacts.

If you only need day-to-day guidance, start with the artifact consumption guide. This page is the short reference for what counts as a durable public output family.

flowchart LR
    A[Need an output] --> B[Human review]
    A --> C[Automation]
    A --> D[Multi-file workflow bundle]
    B --> E[HTML and SVG]
    C --> F[JSON and TSV]
    D --> G[Manifest plus linked files]

Main Public Artifact Families

Common public output families include:

  • SVG tree figures
  • HTML reports
  • JSON summaries
  • bundle directories with manifests and supporting files
  • reviewer-facing TSV ledgers

Stable Families Worth Depending On

The most important schema-governed public families include:

  • FASTA-to-tree model tables such as prefix.model.tsv
  • FASTA-to-tree support tables such as prefix.support.tsv
  • clade review tables such as clade-table.tsv
  • branch review tables such as host-switch-branches.tsv
  • derived comparative trait tables such as comparative-traits.tsv
  • comparative model-output tables such as comparative-summary.tsv
  • event ledgers such as biogeography/event-table.tsv
  • workflow and report manifests such as prefix.manifest.json and comparative-report.manifest.json

When the documentation names one family here, downstream code should depend on its stable headers or top-level keys instead of scraping one ad hoc example.

How To Read These Outputs

  • Use HTML and SVG outputs for human review.
  • Use JSON and TSV outputs for downstream automation.
  • Use manifests when one workflow emits multiple linked files.
  • Use benchmark and parity summaries through their documented output surfaces rather than inferring structure from one fixture or screenshot.

What This Page Does Not Promise

  • It does not claim that every output file ever written by the repository is a stable public contract.
  • It does not claim that every artifact is equally important for trust or publication review.
  • It does not turn one output schema into proof that the underlying science is already closed.

These files matter because they let you inspect the result instead of relying on a one-line success message.