Backtest Anything You Can Describe — Claude Code + Real CME Tick Data

Three years ago, building your own backtesting stack meant being a developer. That's over. In this video I start from nothing and build a Rust backtesting engine on CME tick data — then a live tick chart from a websocket, alerts into Telegram and Discord, and finally I hand an AI a research paper as a PDF and say "backtest this." No script, no cuts, no edits. 46 minutes, every step reproducible. The point isn't the specific tools. Whatever strategy you can describe in plain language, you can now test on real tick data. Most people sitting on a decent idea assume it's out of reach. In 2026 it isn't. On history: I use the 5-year package, not the longest one available. Market structure changes — backtesting on 10-year-old data means testing on a different economy. More history isn't automatically better history. Two things worth staying for: a strategy with a 66.5% win rate that still lost $3,253 over 224 trades (24:39), and why I don't think order flow is an edge after running roughly 200 strategies on it (40:16). Everything here runs on a $79 tickstream subscription and a €15 Claude plan. No proprietary platform, no licence fees, no black box. ⚠️ Results shown in this video come from a 7-day dataset. That is far too short to conclude anything about a strategy — no walk-forward, no out-of-sample split, no Monte Carlo. This is a setup tutorial, not a performance claim. ━━━━━━━━━━━━━━━━━━━━━━ CHAPTERS 0:00 What you'll build (and why I set up a fresh Mac from zero) 1:25 Why your own engine beats MetaTrader 5 and QuantPad 3:21 Step 1: which tickstream subscription you actually need 4:54 The dashboard: where your API key lives 5:44 Step 2: Visual Studio Code 6:14 Step 3: why Claude Code over GPT, Cursor and orchestration 7:30 Pro vs Max plan — and Opus 5 vs Fable 5 explained 9:36 Installing VS Code 10:30 Installing the Claude Code extension 11:22 Logging in and switching the model to Opus 5 12:13 The Ponytail skill: cutting your token costs 14:11 Project setup: turning the docs into tickstream-docs.md 16:24 Adding my "How to Code a Trading Algorithm" roadmap 18:04 Storing your API key in a .env file 19:04 /ponytail ultra and the bypass-permissions warning 20:35 The master prompt: building a Rust backtesting engine 23:55 What Claude built: NQ.bin and the Rust engine 24:39 First real backtest: VWAP 2nd deviation band 26:32 Prompt 2: a live tick chart from the websocket 27:50 Why this replaces the tools the gurus sell you 29:07 Build your own platform: Telegram, Discord and email alerts 32:33 Claude finding and fixing its own bugs 34:12 Running the live chart: real ticks and big prints 36:29 Adding a 5-minute timeframe on the fly 37:44 Backtesting an SSRN research paper straight from the PDF 40:16 The secret: order flow is NOT an edge — options are 41:25 Which options package you actually need 42:57 Tip: use Parquet instead of .bin 44:19 The paper's verdict and an Instagram-ready equity chart 45:25 Wrap-up, links and support ━━━━━━━━━━━━━━━━━━━━━━ EVERYTHING USED IN THIS VIDEO Links can get inside Blogarticel I showed at 16:24 ━━━━━━━━━━━━━━━━━━━━━━ Storage warning: pulling five years of NQ ticks needs 100–200 GB free. Want the Discord webhook build — broadcasting CVD divergences to your community in realtime? Leave a comment and I'll record it. Stuck on the setup? Comment, or reach us on Discord, X or Instagram. Happy to help with VS Code, Claude Code, data or API config. #algotrading #backtesting #claudecode #cmedata #quant #tickdata #rust