experiment-workflowlisted

# 1. Introduction: What is this? A comprehensive, canonical testing workflow—for any system where precision is required. The goal is to RIGOROUSLY evaluate systems by isolating variables (ablation studies), integration, regression, unit-tests, and presenting raw metrics without black-box assumptions. # 2. Core Principles - **Variable Isolation:** Test one specific condition at a time (e.g., one file format like `.pdf`, one document size, or one specific chunking strategy). - **The Engine (THE AI) and the Architect (THE USER):** The agent acts purely as an instrument of the architect's will. It drafts tests and executes commands, but halts at every critical junction for the creator to evaluate and steer. # 3. Experiment Structure (The Blueprint) Every experiment logged or run using this workflow MUST include the following components: ## A. Objectives (The Questions We Want Answers To) Before running tests, user must clearly define the exact hypothesis or goal; the questions he wants answers to, and whether the experiments will produce such answers? Test no 1: This is an example of one of the RAG experiments in which we aimed to test each file separately. - Does the system handle academic documents better than narrative books? - At what document length does context retrieval begin to degrade? - Does altering the chunk size from improve semantic matching without losing keyword precision? ## B. The Source Material Test no 1: - Define the exact absolute path to the test file(