ticket #000039: close — Phase 0 + Phase 1 landed; Phase 2 split to #000050
#000039 (the optional sqlite-vec retrieval backend) closed 2026-05-12.
Shipped: Phase 0 doc + Phase 1 — arborist/search/vec.py (VecBackend,
chunk_vecs vec0 + vec_meta sibling tables, embed_documents
incremental/--rebuild, pluggable Embedder w/ fastembed bge-small-en-v1.5
default), CLI (arborist embed [--limit/--batch-size/--quant/--rebuild]
+ search --backend vec + ingest --embed eager opt-in), [vec] extra,
--quant {float32,int8} with the int8 head-to-head (3.8-4x smaller,
recall ~= float32 — int8 is the production config), ingest integration
+ idempotency (§14), embed-throughput measurement (§14.6 — ~4 chunks/s
contended, ~2.4 GB int8 full-corpus, full backfill abandoned as a
days-long batch job, non-vec ingest unchanged), 16 vec tests.
Demonstrated on crawl_appliedcombinatorics_org.db + a 54K-chunk partial
on wiki shard 000. UNGROUNDED hits, never proof path; vec config folds
into governance_policy_hash (noted in Phase 1, wired in Phase 2).
Phase 2 (RRF hybrid fusion in query.py — wire VecBackend as a 5th
retrieval route alongside the 4 FTS5 routes) → new ticket #000050,
gated on (a) a corpus backfill and (b) a recall bench clearing the
5pp floor. New doc-only scaffold docs/tickets/ticket-000050-vec-rrf-
hybrid-fusion.md (the design already lives in #000039 §4.2 + §8;
#000050 is the tracked continuation). Next ID 000050 -> 000051.
TICKETS.md: #000039 row -> closed, #000050 row added.
(Working tree also has parallel-clone work — ticket-000048-*.md
modified, ticket-000049-*.md untracked — not touched here.)
russell@unturf.com
906606072d