Curate and share our evidence

Our planned session

We spend 45 minutes curating ten question/source pairs, evaluator comparison, and regression result. We publish only reviewed results and label unfinished work.

Weekly outcome

A small RAG evaluation report calibrated against our own labels.

Readiness check

We build on Week 29: A consolidated RAG baseline with at most one justified retrieval improvement. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Human approval gate

We inspect disagreements and never substitute an automated score for source verification.

View Week 30: agent roles, decisions, and evidence →