db: prefer bucket-mounted /data/lineage.db with baked-in fallback

ux: additive filter (click=narrow, more=add) + hub-link icons on every dataset row

explorer: clickable legend filter (toggle relationship types in map + ego views)

explorer: adaptive layout — force for hubs (high fan-out), hierarchical for chains

v3: LLM-judge consensus labels (vLLM Qwen3.6-35B-A3B on 18k edges)

v2 classifier: audit-driven type fixes (schema-check exact_copy, lang-hint translations, demote name-only cleaned, categorical filter)

explorer: overall map default (hubs+chains, sized by derivatives) + corner colour legend

analysis dashboard (Overview tab) + Explorer; type/copy/translation/producer breakdowns

perf: no child cap; hierarchical no-physics layout, straight edges, selective labels, gzip

Tufte restyle + examples dropdown + lru_cache/graph-cap/cache-headers perf

scale: +offset-45k slice

scale: merged 45k parent-seeded run (~11k inferred edges)

fix: drop self-loops + break mutual cycles (DAG)

enrich: 19k-partial lineage.db

full 8k-run lineage.db (1860 inferred edges)

v0: partial-data lineage explorer (FastAPI + vis-network)

initial commit