Bug

gbrain sync --source-id <named-source> --full enumerates files from the source's local_path correctly but writes the resulting pages to the default source instead of the named one. The named source remains at 0 pages after sync.

Repro

gbrain sources add gstack-code-foo --path ~/dev/foo --name "foo"
gbrain sources federate gstack-code-foo
gbrain config set sync.repo_path ~/dev/foo
gbrain sync --source-id gstack-code-foo --full
gbrain sources list
# Expected: gstack-code-foo shows N pages
# Actual:   default jumps by N pages, gstack-code-foo stays at 0

Verified empirically: synced 12 different repos with named source IDs, all 12 named sources still show 0 pages while default accumulated 1,777 pages total.

Why this matters

Per-source filtering on gbrain query --source <name> becomes impossible — everything lives in default. Use cases like "search only my code, not my notes" or "exclude this client's repo from cross-search" can't be expressed.

It also defeats the per-repo cleanup story: gbrain sources remove gstack-code-foo --yes cascade-deletes 0 rows since nothing ever went there, while the actual data sits orphaned in default.

Workaround

Set global sync.repo_path per-repo before each sync (which is what I did to get the indexing done), and accept that everything lives in default. Loses per-source filtering but at least the data is searchable.

Environment

gbrain 0.32.0, PGLite engine.