Search past market news to practise human forecasts or supply date-limited evidence to an AI-agent backtest.
Building an auditable gamma-exposure study, aligning option positioning with the next trading day, and learning from a clean null result.
How I built a restart-safe Hyperliquid candle collector around a 5,000-candle history limit, a shared IP rate budget, and ClickHouse-backed recovery.
A read-only Interactive Brokers research system that compares Client Portal Web API and TWS socket API coverage, endpoint types, output schemas, and practical discovery limits.
I moved IvyDB option prices into ClickHouse, measured compression by column, and chose codecs from full-table storage results.
SSH hardening, a sandboxed AI agent user, UFW, and per-user WireGuard routing. This is the setup I use to run OpenClaw on a VPS.
I screen for liquid ETFs with long histories, select a low-correlation basket around two anchors, and set its weights with Hierarchical Risk Parity.
I compare two models for correlated losses in a BBB bond portfolio and show where their tail-risk estimates diverge.
A minute-level crypto futures pipeline built with CoinMetrics and QuestDB, covering the schema, ingestion logic, and what it took to store 480 million candles.
A difference-in-differences test of how the Ethereum Merge changed the relationship between futures basis and volatility, with Bitcoin as the control.
