Retrieval pipelines
The chain that turns a query into model context — chunking, embedding, indexing, retrieval, reranking and assembly — determining what evidence a model sees before it produces an answer.
The position
The retrieval layer is where financial AI actually fails, and it is the least governed component in the stack. Model risk management scrutinises the model and waves through the pipeline that decides what the model is allowed to know. A reranker change alters outputs as surely as a weight update, ships without review, and leaves no trace in any model inventory. Retrieval configuration belongs under change control, and virtually nowhere is it.
This section is a claim, not a definition. It is argued rather than asserted, and it is the part of this page you are invited to disagree with. Corrections and counter-arguments to [email protected] are published.