Indexing the corpus

PROTRAILBLAZER

Reranking · Two-Stage Retrieval · Flow
RETRIEVED0
RERANKED0
LATENCY0ms
REL@K
TOP-5 Δ0
TOKENS READ0
LATENCY + TOKENS SIMULATED · REL HAND-LABELED
Query presets
Pipeline
Final K
Teaching simulation

Scores here are hand-authored features, not a real model. Real rerankers are cross-encoders, late-interaction models, LLM judges, or learned rankers. Latency and quality depend on model, hardware, batching, and corpus. Tap any candidate row for its feature breakdown.