Critical examinations of statistical methods, data architecture decisions, and quantitative finance — from a practitioner's perspective. When textbook theory meets production reality.
Cloud platform commitment is the largest architectural decision most teams make, and the one most often made invisibly. Why "best service for the job" produces a system whose exit cost the vendor sets — and what regulatory pressure on egress fees in 2024 quietly admitted.
Every proprietary dependency is a future rewrite. Why open source first is an infrastructure decision, not an ideology — and why exit cost is the system property nobody designs but everyone eventually pays for.
AI makes custom tooling accessible to anyone who understands the underlying systems. The value of knowing how things work keeps going up, not down — because that knowledge is now load-bearing for using the tool at all.
PCA finds directions of maximum variance — but variance isn't signal. When financial data violates every assumption PCA makes, that "90% explained" might be 90% noise. A practitioner's guide to when dimensionality reduction lies to you, with Marchenko-Pastur as the principled noise floor.
Engineering labor has migrated from writing code to specifying what code must be. Why context, not prompting, is the discipline that bounds AI-assisted development — and why the generated code is not the asset, the specification is.
When co-movement isn't equilibrium. Why correlation lies on non-stationary data, how cointegration tests rescue the analysis, and why the half-life of mean reversion sets the practical bound on what's tradable.
The right question isn't which time-series database is better — it's whether you want a relational engine that handles time-series, or a time-series engine that handles relations awkwardly. Why TimescaleDB beats InfluxDB for trading workloads: SQL composability, cardinality, joins, and the operational tax of splitting your storage.
Using the Marchenko-Pastur distribution to separate signal from noise in correlation matrices. What eigenvalue decomposition actually tells you about financial data.
Why I chose bare metal over cloud. Storage, networking, the ops overhead nobody warns you about, and why it's still worth it.
Why a strategy with 60% win rate still loses money. Slippage, fees, and the reality of high-frequency signals.
Regime-dependent strategies, the stationarity trap, and survivorship bias in crypto backtesting.
Operational complexity, state management, and when each one actually makes sense.
Performance per watt, cost comparison, and practical experiences running analytics on ARM.
What momentum indicators actually measure, why "overbought" doesn't mean what you think, and empirical evidence from 431 pairs.