What this case covers
Dongqing turned Xiaohongshu virtual materials product selection experience into Sweet Potato Radar, continuously recording product sales and connecting to MCP for AI analysis. The article combines a real product selection library to discuss sales changes, abnormal data, and human judgment, and also provides development requirements that can be handed to Codex.
Questions it answers
- Why is cumulative sales insufficient to judge an opportunity?
- How can you tell whether a product is starting to scale?
- How does collection failure interfere with sales judgment?
- How should candidate products be ranked for observation?
- Beyond the tool score, what else still needs to be judged?
Our observations
- If data collection anomalies are treated as zero sales, they will create false growth signals.
- The ranking is suitable for arranging observation order; whether it can be delivered still requires the operator's judgment.
From the author
I built Sweet Potato Radar not to let a score place bets for me, but to turn what was originally vague experience into data that can be checked.