Infrastructure and methodology articles for systematic traders
Risk management lessons from an AI-fund drawdown: leverage, volatility, short track records, position sizing and decision bias. Includes a precommitted drawdown-review framework.
Most stacks tune a screener then a separate strategy. KreamEdge evaluates both in one end-to-end loop with a shared signal engine, portfolio-level fitness, and a shortlist-then-joint search so selection and trade management stop drifting apart.
A flat dollar-volume liquidity filter behaves differently in 2010 than in 2026, silently narrowing long backtests to mega-cap names and biasing the universe. This note quantifies the distortion and shows how a liquidity deflator corrects it.
A walkthrough of building a production Server-Sent Events (SSE) feed on PostgreSQL: the append-only event log pattern, a migration from sequences to a gapless commit-ordered counter, five hardening fixes, and the consumer failure spectrum. Infrastructure methodology, not hype.
Comparing fixed fractional, Kelly and sigma-targeting position sizing on a multi-version KreamEdge backtest bench. The bench cannot cleanly isolate a sizing edge.
We built a 3-state SPX percent-rank market-regime gate and tested it on 14 systematic quant backtests. None improved. The base rates and the structural reasons why – plus what to try next.
A critical reading of the TradingAgents multi-agent LLM trading paper, explaining why backtesting LLM-based trading agents is fundamentally problematic and why most published results should be treated with extreme caution.
Disclaimer. Content on KreamEdge is published for informational and educational purposes only. Backtest results and strategy outputs do not constitute personalised investment advice nor an invitation to buy or sell any financial instrument. Past performance is not indicative of future results. Financial markets involve risk of partial or total capital loss. Each reader is solely responsible for their investment decisions and risk management. Full bilingual disclaimer on the community page.