Blog

Automated feature generation (tsfresh, Featuretools), budget-aware model search (FLAML's cost-frugal optimizer), and the WorldQuant formulaic alpha factory — the parts of the research pipeline this blog's search-and-overfit arc never covered, and what the arc's own results say about them.

Every leveraged position advertises the price at which it must be sold. How to build the on-chain liquidation depth chart and the CEX liquidation heatmap, model cascade dynamics as a reproduction number, and trade forced flow as a signal instead of only fearing it as a risk.

An active v3 LP is running a market-making book with gas costs and no cancel button. Size ranges from a GARCH vol forecast, frame rebalancing as fee-gain vs realized-cost, derive the position delta from tick math and hedge it with perps, and understand JIT liquidity and the honest pitfalls of backtesting LP from on-chain data.

MEV stopped being a lone bot and became an assembly line: searcher, builder, relay, proposer. A technical walk through proposer-builder separation, MEV-Boost's sealed-bid block auction, why the searcher's margin gets bid away, order-flow auctions as the new moat, and how Solana's Jito model differs.

Finding an on-chain arbitrage cycle is the easy half. The hard half is winning the sealed-bid priority auction against every other searcher who found the same cycle. A technical breakdown of atomicity, flash loans, backrunning, and where arb profit actually survives — including non-atomic CEX-DEX.

Impermanent loss, concentrated-liquidity leverage, and loss-versus-rebalancing derived from first principles. The closed forms, the σ²/8 result, per-block markout, and the exact fee-versus-volatility condition under which an LP actually beats holding.

Almgren-Chriss and VWAP only set slice budgets — execution is won in the tactics layer. Escalation timers, maker-taker break-even math, amend vs cancel-replace queue semantics on Binance/OKX/CME, iceberg anti-signaling, and a Python per-slice state machine.