Bless dn-vector-membership-store: panel-hardened, measured, graduation-shaped — one owner signal unblocks four tracks

[ascalva]

The ask (self-contained)

docs/design-notes/vector-membership-store.md (dn-vector-membership-store — one vector row per idea-atom + a membership relation owning occupancy/currency; the store model finding-0168's five owner rulings mandated) is design-complete and awaiting only your blessing signal. It landed on main by direct commit on 2026-07-23 and was revised 2026-07-27 after its three-seat adversarial panel (core BLOCK · math BLOCK · systems RATIFY-WITH-AMENDMENTS — all 16 findings dispositioned, every empirical figure re-derived from primary sources). It predates the merge-gated regime by one day, so it never had a blessing event: status: draft, and the old hand-flip ceremony no longer exists.

Proposed path (rec): you answer this issue with the blessing (your closing comment IS the ruling, citable as #N), the orchestrator then runs /graduate to mint the two session-plans the note itself sketches (§8: store+land+read · rebuild+gauges+probe+compaction) plus the agent-drafted amendment carrying the two §6 re-home banners into the ratified parent dn-temporal-code-corpus — all in one PR whose merge is your hand, per the regime.

Why now — this note is the named blocker of four tracks

• erratum-relation: bp-129 PD-A and bp-131 PD-H both park with re-entry "dn-vector-membership-store reaches ratified" (verbatim).
• ops: tracks/ops.md — "structurally retires the re-land idiom OPS-1 is bounding"; finding-0169 (resolved, bp-103) closes with "under that model there is no re-land at all, so this class of defect cannot recur."
• code-ingest: the wedged backfill's economics — 52,755 duplicated embeds vs 22,502 atoms, 2.34× measured waste; the note forbids ever re-running the old backfill.
• deployed-instances (PR design(clones): the amnesiac clone — structure ships, the corpus stays home (draft) #26): the structure seed's natural substrate is atoms+memberships, and §D6's n(v) histograms ARE the L0 letter content.
• Plus the rename-stable-identity gap (supersession-lifecycle §7; your own oq-0019 recollection) — closed structurally by D0 + addendum 4.

Two sub-confirmations bundled (say "confirmed" or override)

1. S4 (lance bloat) and F4 (file-grain revert) — the panel's two figures the note disputes rather than closes (245 MB disk vs ~232 MB raw payload; adjacent-dedup preserves [A,B,A]). The revision's re-measurements are in the note; your confirmation removes the residual doubt.
2. Graduation may pin one small residue the note leaves open: embed-reuse keyed to the embedder pin (a model change must invalidate every reuse — the ops-brainstorm "have I seen this?" caveat; same embedder-pin concept as dn-amnesiac-clones). One line in plan A's interfaces.

What does NOT wait on you

Nothing else. Plan A (store+land+read) has zero operational preconditions. Plan B (rebuild) is sequenced behind the wedge-clearing you already own as an op (D7/S6) and the first successful commit_diffs capture (step 0 — never yet run; job 300240 died in TimeoutError).

Glossary

• finding-0168 — docs/findings/ (frozen): "vectors are the versioned entities" + 4 addenda (fork semantics, append-only, n(v) first-class, rename-mints-no-vector).
• finding-0169/bp-103 — the re-land cost defect (11.7s/materialization, backfill death at 847/1,542) and its landed zero-marshal fix.
• D0 — atom identity = header-free canonical body (your 2026-07-27 ruling; rename mints 0 atoms vs 38).
• oq-0019 — your rename-threshold recollection, answered: membership is the successor mechanism.

[ascalva]

Resolved by the owner's merge of PR #29 (2026-08-06) — under the merge-gated regime the merge IS the blessing, so dn-vector-membership-store is ratified and bp-151..153 are readied by that act.

Both bundled sub-confirmations were answered the same day: S4/F4 re-measurements stand, and embed-reuse is pinned to the embedder identity (EmbeddingConfig.model + dim), carried in bp-152 §6 with its own test case.

Closing late as bookkeeping: PR #29's body carried the closing keyword inside a code span (closes #27), which GitHub does not parse, so auto-close never fired. Issues #31 and #34 closed correctly because their commit messages carried the keyword unbackticked. Noted so the habit gets fixed rather than repeated.

Downstream since: bp-151 merged (PR #32, rename cost 11,096 → 3), Amendment A1 + bp-155 merged (PR #33/#36, → 0), Amendment A2 renamed the occupancy coordinates (PR #35). bp-152 is next.