Compression can make graph links larger than vector payloads

When compressing vectors for graph search, budget adjacency storage and traversal state separately from vector payloads. Compression can make link lists larger than the vectors, so reducing vector bytes no longer yields proportional index capacity or traversal speed. Derive capacity from the combined record, then reserve space for upper levels, identifiers, visited state and candidate queues. For an accelerator, benchmark the packed similarity primitive as well: a compact bit representation does not imply efficient native population-count execution. This is a storage and execution qualification rule, not evidence that a particular graph or device is faster.

Applies when: Quantized vectors are used in a graph nearest-neighbor index. Keywords: ternary vectors graph adjacency memory packing similarity accelerator