Verify + land #537 reuse-under-drift (ADR-202 WIN) — already productionized

Productionize #537 — reuse-under-drift (ADR-202 WIN)

Assessment: #537 already arrived productionized — unlike #479/#506 (nightly PoCs I built out across 6 steps), #537 is a mature "prove-not-hype" PR that already did the work:

• ✅ Implement — crates/ruvector-diskann/src/reuse.rs (448 lines): DriftingIndex + RecallTrigger with a RebuildPolicy enum, wired into ruvector-diskann behind a feature flag.
• ✅ Test — 8 reuse-specific unit tests in reuse.rs (part of the 17 diskann tests, all green).
• ✅ Benchmark / validate — examples/diskann_real_trajectory.rs + triggered_rebuild.rs on a genuine learned-GNN trajectory (contrastive link-prediction over ogbn-arxiv). ADR-202 WIN: policy P k=4 holds ~98.7% recall@10 at near-zero cumulative rebuild cost (vs 246s / 30 builds for always-rebuild), at n=20k and n=50k.
• ✅ Secure / rigor — pre-registered + frozen gate before any contender run (docs/plans/bet1-productionize/PRE-REGISTRATION.md).

Loop verification (this issue)

• Merged PR #537 onto current main cleanly (+2183/-0, purely additive).
• cargo test -p ruvector-diskann → 17/17 pass.

Recommendation

Land the existing PR #537 — it's complete and rigorously validated; re-deriving it would be redundant. (No competing PR opened.) The loop adds verification, not a rebuild.

Source: #537 · ADR-202. Loop job 05bbd4fb.

Inspect PR #537 and the implementation in crates/ruvector-diskann/src/reuse.rs, including its reuse-specific tests. Run cargo test -p ruvector-diskann to confirm the stated 17 tests pass; done means verifying and landing the existing PR rather than reimplementing it.