Purpose
The lineage DAG viewer shows how ModelBreeder wants readers to think about descendants. A child model is not just “new weights.” It has parents, an operator, evidence, resource profile, approval state, rollback target, and lifecycle state.
Sample lineage DAG
Choose a sample ecology and inspect parentage, operators, evaluation states, and retirement outcomes.
Minimum lineage record
STRUCT LineageNode
artifact_digest
parents
operator
operator_config_digest
eval_suite_digest
scorecard_digest
resource_profile_digest
approval_record
rollback_target
lifecycle_state
END STRUCTThe viewer uses sample data. A production implementation would read signed lineage records from the package registry.
How to use it
Select a sample lineage and click each node. The viewer shows the artifact digest placeholder, operator, lifecycle state, rollback target, and evidence bundle that should exist before any descendant is promoted. The point is not the visual itself; the point is the discipline that every active model or adapter should be explainable through ancestry, operator configuration, evaluation evidence, resource profile, owner, and retirement state.
Use this pattern when designing a file-backed registry: store the lineage as JSON, keep the large artifacts separate, hash every source package, and never delete retired nodes merely because they left active routing.
Source reports used for this guide
These reports are preserved verbatim in the site archive. The guide above is an editorial synthesis and may narrow, qualify, or reorganize claims from the source material.