Crawler now records the deduped cross-domain link graph (edges table).
recompute.js runs real iterative PageRank over that graph, blends it with
accumulated in-degree + the Majestic seed prior, and patches every doc's
`score` (Meili's authority tie-breaker) via partial updates — preserving
titles/bodies/embeddings and triggering no re-embedding. systemd timer runs
it every 3h so ranking sharpens as the 1.5M-page crawl densifies the graph.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
