Evidence First
Build a quant trading strategy with Claude, from a trading article to a live bot, without fooling yourself.
Every idea goes through the same eight steps.
One real build, start to finish. A TradingView article on MACD and CVD became mechanical rules, a Rust engine and a backtest. Then came two pre-registered studies, and both failed. Neither changed the live bot. That's the method working, and it's what you'll learn to do with your own ideas.
- 01Idea
- 02Rules
- 03Data
- 04Backtest
- 05Study
- 06Port
- 07Demo
- 08Review
What's inside.
Start here
Draw the whole loop and name the evidence each step produces.
- 0.1The loop on one pageThe tickshift build in eight stepsComing soon
- 0.2Claude as your quant pairA project file that fixes the report format and the safety rulesComing soon
From idea to rules
Turn a discretionary article into rules a computer can run, with every assumption labeled.
- 1.1Turn an article into rules you can test“From the article” vs “Our choice”Free
- 1.2The four-action ruleReversals as two signals on one barMembers
- 1.3Define indicators exactlyWhy the CVD slope is delta over notional across six barsComing soon
Data you can trust
Download, cache and check exchange data, and spot data that leaks the future.
- 2.1Public exchange data, in UTC with half-open rangesThe Bybit V5 market endpointsComing soon
- 2.2Building CVD from tick trades5-minute bars with delta and notionalComing soon
- 2.3Hindsight trapsZones are hindsight, delayed by their lookbackComing soon
A backtest that could have been traded
Build and read a perp backtest that includes every cost and every way to lose.
- 3.1Next-bar fills, fees, slippage and fundingTaker fees, slippage and slippage stressComing soon
- 3.2Leverage, ROI stops and liquidationA −6% ROI stop at 10× sits 0.6% from entryComing soon
- 3.3Reading a backtestWhy fewer trades can look better than they areComing soon
Not fooling yourself
Design a study whose answer you'll trust, and accept a no.
- 4.1Train and test periods, named with their datesOne train year and two test yearsComing soon
- 4.2Pre-registrationThe flat-cross plan, written before any variant ranComing soon
- 4.3Multiple testingJudging the best of a 226-variant sweepComing soon
- 4.4Controls: placebos and the bootstrap500 permuted-timing seedsComing soon
- 4.5Negative resultsTwo studies that failed and left the bot unchangedComing soon
The research note
Publish a result a skeptic can check trade by trade.
- 5.1Verdict first: the backtest noteQuestion, verdict, results by periodComing soon
- 5.2Look at every tradeTrade index and trade cardsComing soon
From backtest to live
Ship a strategy that provably trades like its backtest, with limits that stop it before it hurts.
- 6.1Parity tests with fixturesThe live engine must make exactly the backtest's tradesComing soon
- 6.2Demo first, and record every decisionDemo venues and a decisions tableComing soon
- 6.3Risk limitsA halt at a 10% loss of the fundComing soon
- 6.4Alerts with chartsSeeing what the model missedComing soon
Capstone
Your own idea through the full loop, with Claude doing the work and you checking it.
- 7.1Write your idea's rulesEvery assumption labeledComing soon
- 7.2Pre-register, run and publishWhatever the verdictComing soon
- 7.3Ship it to demo, or write down what you learnedParity firstComing soon
Try the first lesson. It's free.
Ten minutes, two quizzes, and you'll know how to read any trading article like a backtest.
Education only, not financial advice. No result in this course predicts future results.