Now I understand the shape. Pages have no wikilinks/backlinks in the body — they're raw archives — so the

extract

phase finds nothing to link (0 links every cycle), hence near-total orphans. And

synthesize

(the phase that would generate connective/gap-filling pages) is off because

session_corpus_dir

is unset. The honest situation: enabling meaningful cross-linking here isn't a single config flip — it needs a synthesis corpus dir configured and likely embeddings-driven relation extraction, which changes brain behavior materially. Let me check what setting

session_corpus_dir

would even point at, and whether embeddings are actually populated (retrieval quality is the real thing that fills gaps at query time).