LAB

RAG vs Long-Context Hallucination & Retrieval Lab

Vector topology, context decay & needle-in-a-haystack stress simulator
Default Hydrated
Top-K Recall Score
0.84
Target chunk retrieval accuracy
Hallucination Risk
0.12
Predicted probability of false generation
Lost-in-Middle Bias
0.22
Positional attention attenuation
Latency (ms)
142ms
End-to-end processing estimate
Cost / 1k Queries
$0.38
Embedding + Generation token pricing
Vector Space Topology & Cluster Attention
Nodes = Chunks, Star = Query Embedding
Lost-in-the-Middle Positional Attention
Cost vs Recall Pareto Frontier
Enjoy this tool? Build your own with Super