From data you can trust to a system that runs on its own
Advanced · Data and production · with code
Level 1: Where they come from and what they carry inside
Before the first line of strategy: who touches a price before it reaches your disk and what they charge you for it, what exactly the number you call the "close" means and why two serious providers can hand you two different ones on the same day, why every fundamental datum has two dates and using the wrong one inflates any backtest, and why joining two sources on the ticker silently returns another company's data.
Level 2: The biases that break a backtest
The four mistakes that turn an excellent backtest into a worthless number: the graveyard of companies your data download deleted without telling you, the information from the future that slips into the signal through a single day of misalignment, the decisions about period and universe you take after looking at the result, and the set of runnable checks that separate a credible equity curve from one that only proves how well the past fits itself.
Level 3: Adjustments and corporate actions
Why yesterday's price is not the same today: how a split or a dividend forces the whole historical series to be rewritten backwards and which adjustment factor does it properly, how a price series differs from a total return series and how far apart they drift once dividends are involved, what happens to the series of a company that merges, spins off, changes its symbol or delists, and how to rebuild an adjusted series from scratch and audit it until you are certain that the return you are calculating is the one the shareholder actually put in his pocket.
Level 4: Cleaning and validation
The difference between a datum that is missing and one that never existed, and why filling both in the same way ruins the backtest; how to tell a provider's typo from a real takeover bid without deleting precisely the days that matter; how to turn cleaning into automatic checks that stop the process instead of letting it carry on with rubbish; and why versioning your code is not enough if the provider can rewrite the past under your feet.
Level 5: Storing, updating and scaling
The part of the craft nobody teaches and that decides how many experiments you can run in a day: why CSV is convenient and expensive and what a compressed, partitioned columnar format buys you, how to write an incremental update you can run twice without duplicating anything and what to do the day the provider corrects the past, when a database genuinely helps and when it is complicating things for the fun of it, and the real balance between what storage costs, what downloading it all again costs and the latency your strategy genuinely needs.
Level 6: From data to decision
The level that turns the previous five toolboxes into a dataset you can trust: how to line up daily prices, quarterly fundamentals and revisable macro without letting in a single day of information from the future, what alternative data promises and what it genuinely delivers, how to leave behind a dataset somebody else can audit field by field, and a concrete checklist to apply to any table before believing a backtest.
Level 7: From the notebook to the process
The leap almost nobody makes properly: why a notebook that works twenty times in a row is still not a system, how code is split into modules so that the strategy never touches the broker, where configuration and credentials genuinely live without ending up published in a repository, and why a result you cannot obtain twice in exactly the same way is not a result.
Level 8: State and reliable execution
The level that separates a system that loses money to a silly bug from one that does not: what the state of a trading robot is and why it cannot live in memory, how to write to disk without leaving a half-finished file, how to send an order over a network that drops without ending up buying twice, how to retry with growing backoff without exhausting the broker's rate limit, and how to model the order as a state machine so that a partial fill does not leave you with a position nobody expected.
Level 9: Orchestration and the calendar
Who presses the button when you are not there: the three ways of triggering work and the lock that stops two copies trading at once, why the market calendar has to be a piece of data you query and never a hand-written condition, how the dependencies between downloading, computing, sizing, executing and reconciling are declared so that a signal never comes out of incomplete data, and how a real server behaves: clock on universal time, daylight saving shunting the open around for a couple of weeks a year, automatic start-up after a reboot, and the duty to read the real state before moving a single euro.
Level 10: Knowing what is going on
The difference between a system that trades and a system you can trust: structured logging that lets you reconstruct why you bought each share, metrics that separate the health of the system from the performance of the strategy, alerts that only go off when you have to get out of your chair, and a daily reconciliation against the broker that catches the mismatch before it turns into lost money.
Level 11: When it fails
The level that assumes your system is going to break: the map of the five places a trading system genuinely fails and what the safe response is at each one, the hard limits the program imposes on itself and cannot override, exactly what to do when the process dies with an open position in the market, and how to turn every scare into a new automatic check instead of into a promise to be more careful.
Level 12: Testing, deploying and taking the step
The level that closes the course and the academy: why testing a trading system demands isolating the pure logic, replacing the broker with a double and deliberately provoking the rare cases; the ladder of simulation, paper account, small capital and real capital, with what each rung validates and what each rung lies about; how to deploy without breaking anything, with a rehearsed rollback, state compatible across versions and no configuration changed behind your back; and a concrete checklist before putting real money in, followed by the full map of the nine courses and the warning that is not a formality.