AI - Gold (4) Gold currently has a V4 version, but before diving into V4, I want to talk about another system we’ve been working on: The Gold Research Report System. This actually stems from last week’s auxiliary research after the V1 and V2 agent PNLs crashed. If a simple regime filter doesn’t work, what other options do we have? Can we add an additional research layer on top of the results derived by agents from candlestick charts? For example, using macroeconomic data and leveraging LLMs for subjective analysis (instead of mathematical models) to create a trend filter? Taking it a step further, can we use various macro and technical data to have LLMs directly generate weekly and intraday research reports? So, based on the existing trading agents, we developed this weekly and daily research report system for gold. The core idea is actually the same as the trading agents: Feed data to AI -> Provide AI with research directions and prompts/skills -> See what results it produces. 1. News Agent: Responsible for sorting through news and events. 2. Macro/Flow Agent: Focused on macroeconomics, cross-asset analysis, fund flows (ETFs), and positioning structures. 3. Technical Agent: Handles price structures and key levels. 4. Master Agent: Synthesizes conclusions from the different modules into a mainline judgment. 5. Writer Agent: Converts the findings into readable daily or weekly reports. We also created two versions—one using Claude and the other using GPT—for comparison. After running it for a week, the results were surprisingly good. Below are a few examples: Images 1-3 show the judgments from Tuesday to Thursday. Since the results over the past few days have been promising, we went ahead and developed a self-evaluation agent. This agent is designed to review the day’s market action after the daily COMEX close and assess the accuracy of the day’s research report (Image 4). Now, this daily and weekly report system is running every day… Continuing to validate the results…