What ProBitForge Is Building

ProBitForge is a public research log for building and testing AI-assisted crypto quant systems. The first job of this site is not to sell a signal. The first job is to make the research process visible enough that claims can be checked, risks can be named, and bad ideas can be stopped before they reach live execution.

The system boundary

The research stack is designed around a simple rule: models may help analyze, summarize, and propose, but they do not press the buy button. Execution belongs to controlled trading infrastructure. Research notes, memory retrieval, strategy reviews, and risk logs are supporting layers, not a shortcut around validation.

That boundary matters because crypto markets punish vague confidence. A backtest can look clean because fees were ignored, sample windows were cherry-picked, a regime changed, or a signal was accidentally fitted to history. ProBitForge treats those failure modes as first-class research topics.

What we will publish

• Research notes: strategy hypotheses, signal families, and market-structure observations.
• Backtest lab notes: sample design, walk-forward checks, out-of-sample gates, fee and slippage assumptions, and reasons a candidate was blocked.
• Risk logs: drawdown reviews, false positives, risk filters, and failure patterns.
• System build notes: Freqtrade execution boundaries, OpenClaw-style review loops, Qdrant memory retrieval, and publishing automation.

What we will not publish

• No live account credentials, server credentials, private API tokens, raw account exports, or unde-sanitized operational logs.
• No real-time buy or sell instructions.
• No guaranteed-return language.
• No claim that a good backtest is the same thing as a profitable live strategy.

The operating standard

Every serious strategy note should answer five questions: what was tested, what data was used, what assumptions were made, what would invalidate the result, and what risk gate stops the idea from being promoted too early.

That standard is intentionally conservative. In trading research, a blocked candidate with a clear failure reason is often more valuable than a lucky curve that teaches nothing.

Risk notice

This publication is for research and engineering documentation only. Nothing on ProBitForge is financial advice. Crypto assets are volatile, automated trading can lose money, and historical tests can overfit. Treat every article as a research memo, not an instruction to trade.