Evolution Lab Introductory 1 minute read Updated 2026-06-29 UTC

Lineage DAG Viewer

A guide to visualizing parent-child relationships, adapters, merges, evaluation evidence, and release states in model-breeding labs.

Research statusBrowser-local tool guide Publication statePublished Reviewed byMichael Kappel Source reports3

What the viewer teaches

The Lineage DAG Viewer shows why model breeding needs memory. A descendant should not be a mystery file. It should have visible parents, operators, adapters, merge recipes, evaluation evidence, release state, and rollback target.

A lineage viewer makes capability reusable. Builders can see which branches produced useful specialists, which merge recipes created portable skill, and which archived nodes remain good parents for later experiments.

What to show in every node

FieldMeaning
Artifact digestStable identity for the exact package.
Parent idsThe reusable source lineage.
OperatorFine-tune, adapter, merge, distill, quantize, router mutation, or no-op.
Fitness packetLink to the evidence that gave the node meaning.
Resource profileLatency, memory, energy, and local-fit information.
Lifecycle stateDraft, lab, shadow, canary, specialist, champion, or archive.
Reuse noteWhy this node should or should not become a future parent.
pseudocode
PROCEDURE render_lineage_view(candidate_id)
    node <- LOAD_LINEAGE_NODE(candidate_id)
    parents <- LOAD_PARENT_NODES(node.parent_ids)
    children <- LOAD_CHILD_NODES(candidate_id)
    evidence <- LOAD_FITNESS_PACKET(node.evaluation_packet_uri)
    RETURN DRAW_DAG(parents, node, children, evidence)
END PROCEDURE

Next step

Open the browser-local lineage viewer, then compare it with Lineage DAGs Make Capability Reusable.

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.