Heterogeneous IR benchmark. Phase 2 baselined all 15 public datasets (BM25 avg nDCG@10 0.4261, on the published ~0.43) and classified techniques cross-domain: the BGE hybrid is universal (ARCH-0.5 core), keyword rerankers are domain-conditional, learning-to-rank is portfolio.
Retail Search
ARCH-0.5 · cross-domain core
Phase 2 - Cross-Domain Validation (BEIR)
Test whether Phase 1 techniques transfer across domains. Result: BGE hybrid is universal (the ARCH-0.5 core); keyword rerankers are domain-conditional; learning-to-rank is portfolio.
Livestatus
15 public BEIR datasetsdataset
BM25 avg nDCG@10 0.4261 (on published ~0.43); BGE hybrid universalmetrics
Live cross-domain endpoints
The Phase 2 evidence and live search are public. Learning-to-rank and keyword rerankers stay in the portfolio; the BGE hybrid is the ARCH-0.5 default.
/api/beir/metadatasets, tiers, core architecture
/api/beir/evaluation15-dataset matrix + technique scopes
/api/beir/nfcorpus/searchlive BM25 (Tier 1)
/api/milestones/arch-0.3-bge/searchlive BGE hybrid via runtime embedding