- Published on
An open-source, candle-only crypto pump detector–looking for ML contributors
- Authors
- Name
- DeepHorus
A new open-source tool has been released to help cryptocurrency traders spot pump-and-dump schemes before they collapse. The project, described in a GitHub repository, automatically pulls market data from major exchanges Binance and Bybit, scans for abnormal price movements, and sends real-time alerts through Telegram. Dubbed a 'candle-only' detector, it uses only public market data and requires no exchange API keys.
Pump-and-dump schemes remain a serious problem in crypto, where coordinated groups artificially inflate token prices before selling off. The newly released screener aims to give ordinary traders a defense by analyzing historical and live price action. It runs as a Node.js application built with TypeScript and uses a pnpm monorepo structure, with PM2 for process management in production.
The tool downloads 1-minute and 5-minute market statistics and bulk kline archives from Binance and Bybit, covering a configurable lookback period—five days by default. It assigns each detected event a score, with a minimum threshold of 80 for alerting. The system can also require a 'calm' pre-pump period to reduce false positives.
Users can interact with a live Telegram bot, @pumpdumpscreenerautobot, which automatically subscribes anyone who sends /start and provides commands like /stats and /runs. The repository also includes a web-based manual review interface, allowing users to classify events as wick spikes, weak pumps, sustained moves, volume-only, illiquid noise, or unclear. Each classification can include confidence ratings and comments, and labeled datasets can be exported as JSON or CSV.
The project is particularly notable for its emphasis on creating a human-labeled dataset. By storing annotations separately from raw detector output, the tool enables continuous evaluation and improvement of its screening algorithms. This open approach could appeal to researchers studying market manipulation and developers building more sophisticated detection models.
The creator explicitly invites machine-learning contributors to join the project, suggesting that future versions could benefit from advanced pattern recognition. For now, the detector operates on deterministic rules, but community involvement might introduce adaptive models. The repository provides detailed setup instructions, including database setup with Turso and deployment guidance for Nginx and Let's Encrypt.
While the tool is still in its early stages, it represents a proactive step toward transparency in crypto markets. By leveraging free public data and crowdsourced labeling, it lowers the barrier for traders to protect themselves. As the dataset grows, the project could become a valuable resource for both practical trading and academic research.
