Writing on trading-systems engineering
Short, practical notes on the parts of quantitative and pre-trade infrastructure that decide whether a system stays correct under load: deterministic evaluation, latency budgets, idempotency, freshness, and fail-closed control paths.
The one property that governs every trading system: whether it holds when capital is at stake and load is high. The failures that destroy the most value are silent, not loud.
Read on Substack →One validation engine serves research and production, so a backtest and a live decision can never quietly disagree.
Read on Medium →A pre-trade control that cannot answer inside the latency budget is, in production, the same as no control at all.
Read on Medium →Give each piece of mutable state exactly one writer, and latency stops depending on who else is contending for it.
Read on Medium →Past its freshness deadline a value is absent, not trustworthy, and treating stale as fresh is how silent losses begin.
Read on Medium →An idempotency key makes a retried order place exactly once and never twice, however the network misbehaves.
Read on Medium →One authoritative pre-trade gate beats two that can drift apart and disagree at the worst possible moment.
Read on Medium →When the pre-trade path cannot verify, it must reject rather than pass, because unevaluated is not the same as allowed.
Read on Medium →