I shipped two deterministic, pre-registered benchmarks with JAMES — a local-first Graph-RAG. RAB scores audit replayability against EU AI Act record-keeping articles; LRB scores time-valid retrieval across 4 models × 4 scales. Two preprints + DOI.
Most RAG demos answer "what's the right chunk?" Very few can answer the
two questions a regulator or an auditor will actually ask:
Replay this decision — show me the exact, complete record of how
this answer was produced.
Reconstruct the past — what did your system know at the moment it
RAG has become the default answer for giving LLMs access to private knowledge. And for good reason —...
When an auditor asks "reproduce this exact answer" six months later, you need more than (messages,...
Most "chat with your documents" demos work in an afternoon. Then you hit the last 20%: retrieval that...
How embedding models, chunking, retrieval modes, and query phrasing jointly determine Code-RAG quality.
What RAG architectures are missing between finding information and trusting it.
Standard benchmarks fixate on how well large language model (LLM) agents perform in finance, yet say little about whether they…
