Nine modules. Fundamentals to final project.

Discretionary versus systematic. Why real edge is rare, why most published edges are artifacts, and enough market-structure basics to know what you are actually buying and selling.

From the operation: Why this operation refused to go live for over a year — and what 'no edge yet' costs versus what pretending costs.

Python setup that won't fight you, where market data actually comes from (and what free data quietly gets wrong), and the pandas fundamentals every later module stands on.

From the operation: Build your own data pipeline alongside the lesson — the same shape ours has.

Turning a hunch into a testable, falsifiable statement. What makes a hypothesis honest, and why 'it goes up after it goes down' is not one yet.

From the operation: Pre-registration: writing down what would prove you wrong before you run the test — the single cheapest overfitting defense that exists.

In-sample versus out-of-sample. Walk-forward validation. The overfitting traps that make beginner backtests look brilliant: lookahead, survivorship, costs ignored, parameters tuned on the answer key.

From the operation: Our six-filter promotion gate as a framework — the categories a strategy must clear (risk-adjusted return, consistency, sample size, drawdown, deflation for multiple testing, benchmark dominance). The categories, not our thresholds.

Position sizing that compounds, diversification that is real rather than cosmetic, drawdown control, and why sizing rules matter more than entry rules.

From the operation: A diversified trend-plus-carry book as a worked example of portfolio construction.

How multi-perspective review catches errors a single analyst — human or model — reliably misses. Skeptic, advocate, and risk roles as a repeatable review structure.

From the operation: The boardroom-debate concept from our own stack, generalized into a review pattern you can run with colleagues, checklists, or LLMs.

Paper trading done honestly, broker APIs, order types, and slippage — the gap between the fill your backtest assumed and the fill the market gives you.

From the operation: Why a promotion gate exists between paper and live at all, and what it must check.

Debugging a live system: root-cause analysis, why 'it works in backtest' is the beginning of the argument rather than the end, and the failure modes that only appear with real money attached.

From the operation: Two real incidents from our stack, generalized: a seam bug between subsystems that agreed on everything except the data format, and a feedback loop that silently contaminated its own training signal. Full write-ups in the case studies section.

End to end: hypothesis, backtest, walk-forward validation, paper trade. You finish with a strategy you have honestly tried to kill — which is the only kind worth having.

From the operation: Your project, reviewed against a public rubric mirroring the six-filter gate.