[AdaWorldAPI]

Goal

Research and work-steal from the existing TurboVec9 Lloyd-Max probe (#3), without adding another codec or changing production APIs. Test the learning-adjacent hypothesis: preserving true graph neighbors as candidate edges, not just approximate vector top-k recall.

Reuse

• Existing turbovec9_probe 4-/9-bit scalar index, codebook, upstream deterministic Rotation and exact f32 oracle.
• Explicit independent truth graph: each group of 4 related vectors is labeled before quantization; queries are new noisy observations. No reliance on self-confirming retrieved labels.
• Existing CI workflow, now logs graph-edge-any@k, 4-edge-coverage@k, and f32 top-k recall for f32, scalar-i4, scalar-i9 and shipped-i4.

Falsifiers

• Same graph IDs & seeds for all encoders; fixed noise.
• All old bitplane/codebook tests remain; new graph oracle and uniqueness tests.
• No training, gradient, or NSL quality claim. Graph neighbor retention is a prerequisite only.
• No mixing with ndarray DR512 presence/polarity; 512 TurboVec centroids remain purely numeric.
• No code changes in ndarray or lance-graph, no force-push and no merge.

Run

TV9_WORKLOAD=graph TV9_N=2048 TV9_NQ=64 TV9_DIM=128 TV9_K=10 TV9_NOISE=1.7 \
 cargo run --release -p turbovec --features turbovec9-lab --example turbovec9_probe

See docs/turbovec9-learning-neighbor-probe.md. Converge later at the existing lance-graph-nsl Google-Loss oracle, not via a duplicate learner. Draft until CI and findings review.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-29T14:29:28.486500Z	51c3887	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.