Portable Model Evidence

Embedding model bake-off

Public screening evidence selected static-retrieval-v1 for the full retrieval matrix.

Candidate results

ProfileStatusnDCG@10Warm p95Peak RSSDecision
static-retrieval-v1selected-default0.5910294.63 ms316.52 MiBBest quality-cost Pareto point and promoted to the full public matrix.
code-minilm-l12-v1resource-stop0.52731847.21 ms1.34 GiBStopped after one task because it exceeded both screening budgets and did not beat code MiniLM L6 quality.
code-minilm-l6-v1rejected0.53301155.55 ms1.02 GiBCrossed both laptop screening limits on its first completed task; the partial quality result is not an aggregate claim.
generalrejected0.44661309.51 ms1.00 GiBCrossed both laptop screening limits on its first completed task.
Salesforce/SFR-Embedding-Code-400M_Rexcluded---Non-commercial license and custom architecture are incompatible with the shipped default.
jinaai/jina-code-embeddings-1.5bexcluded---The 1.5B-parameter custom model exceeds portable laptop download, memory, and inference budgets.
nomic-ai/CodeRankEmbedexcluded---Custom NomicBERT architecture is not supported by the portable local inference backend.

Claim boundary

The static profile is the only candidate that completed the screening matrix within the declared laptop budget. Resource-stopped rows are single-task observations, not aggregate quality claims.