Risk Lab

1/ 5R a week. A $5,000 account. Compounded for 3 years. $10.1 million. The same trades with a fixed $50 risk: $44,000. That gap is compounding. Here's the math: what R means, what different R values turn into, and why automated strategies are how you'd actually chase it 🧵

What the crypto market actually does. 98 perpetual-futures coins since 2021, 10 million one-minute bars, six years of open interest and positioning, $4.9 billion of liquidations, all run through the lab's own tape. 1. The day has one loud hour, and every hour has a loud first minute. Every 5-minute bar of the last two years across ten majors, measured against the coin's own typical bar. The quietest slot of the day is 05:55 UTC at 0.7×. Then 13:30 UTC (US data) at 1.6×, and 14:30 UTC, the US cash open, at 1.7×. One 90-minute block carries the day; 03:00 to 08:00 UTC runs at 0.85×. The same shape repeats inside every hour. Minute :00 runs at 1.2× a typical minute, :30 at 1.2×, :15 and :45 at 1.1×, :59 at 0.8×. The clock itself moves price: bar closes, funding settlements, the orders people park on round times. 2. The day's extremes come at the open and in US hours, not at the close. 141,871 coin-days. The day's high prints in the first UTC hour 16.3% of the time, the low 14.0%. A random walk puts about 13% at each end (we simulated it), so the open is only mildly special. The real anomaly is the other end: a random walk puts 13% of highs in the last hour. The tape puts 7.6%. The crypto day does not finish at its extreme. It finishes quiet, and the extremes it prints away from the open come in US hours. Saturday runs at 0.8× a normal day's range, Sunday 0.9×, Monday 1.1×. 3. When the crowd is most long, the next week is the worst one. The largest perp venue publishes its long/short account ratio. 1,908 days of it on BTC since September 2020. Accounts are net long on 79% of days, the median is 1.45 longs per short, and longs paid the funding on 74% of 8-hour settlements over the last 13 months. When that ratio sits in its top decile (2.45 longs per short or more), the next seven days averaged −2.7%, up only 42% of the time. Bottom decile (0.78 or less): +1.7%, up 59%. A typical day: +0.7%, up 52%. The weeks overlap, so call it 25 independent looks per decile. A lean, not a law, but it has leaned the same way for six years. 4. Positions are built all day and forced out in two hours. BTC open interest by hour, 17,743 hourly reads. It grows through every hour from 01:00 to 17:00 UTC, biggest adds at 14:00 and 15:00, and shrinks from 18:00 to 23:00, with −16 and −23 basis points an hour at 21:00 and 22:00. Positions open through Asia and Europe and close out of the US afternoon. Liquidations are the mirror. 1.5 million forced closes on two venues over the last two months, $4.9 billion. The 15:00 UTC hour alone carries 13.2% of every dollar liquidated; 13:00 through 15:00 carry 29%. The quietest hour, 06:00, carries 1.7%. Longs were 55% of the money flushed.

Testing automated trading through Hyper liquid as the next venue. So far, so good.

When you are in the lab, you will find yourself blown away at every turn.

A serious crypto trader's stack, at the tiers people actually buy. List prices, monthly: TradingView Expert · charts · $199.95 Bookmap Global+ · order flow · $99 TraderSync Elite · journal · $79.95 TrendSpider Advanced · backtesting, alerts · $349 3Commas Expert · bots · $110 $838 a month. And the expensive part isn't the subscriptions. It's that none of them know your rule, your journal, or each other. Your chart can't test what you drew. Your journal can't tell you which rule is bleeding. Your backtester has no idea what happened after you went live. Your bot runs a grid, not your edge. Risk Lab is $119 a month. Less than the chart plan alone. And it's better at each job: HOW IT STACKS UP • Charts: the backtester lives on the chart. Draw a level, test it, arm it, same screen. Auto levels, auto fibs, wave counts, TPO, Wyckoff, a 5-second footprint. • Order flow: our own 24/7 crypto tape, not a feed we rent. Footprint, delta, open interest, liquidations, funding. Alerts fire on flow conditions, not just price. • Journal: it measures, it doesn't just log. Heat and run on every trade from real candles, your stops replayed, rule cost in R, a bias lab, Monte Carlo. 20+ importers including prop firms. • Backtesting: honest by construction. Plain-language recipes, 5 years of 5-minute history on 780+ pairs, pessimistic fills, nothing optimistic. • Automation: your tested rule, executed. Stop attached at the exchange and re-checked. Runs on your machine with your keys. Paper first, always. NOTHING TO COMPARE IT TO No other product has a column for these. • One rule, one definition, runs backtest → paper → live. No rewrite between stages. • A luck gate on every test: out-of-sample verdict, attempts counted, sealed. • The forward record sits beside the claim. The strategy seals itself or it doesn't. • Plans matched to trades. Adherence and rule cost measured, not assumed. • Heat and run on every trade you ever took, then your stops and exits replayed. • Theses on a clock: prices stamped at post, sealed, scored by the tape. • Groups with shared strategies and every member's forward record. • Plug in your own model: it reads your journal and runs your tests through the lab. • A Lab Assistant that knows every tool, and a manual with a recipe for every term. • An export we can't read? We build the importer. WHAT TRADING ALREADY COSTS YOU One fitted strategy, traded 100 times before anyone checked it for luck: about 17R. One prop evaluation: $100 to $1,000, paid again every time an unmeasured rule breaks. The exchange's cut: roughly 0.1% a round trip, win or lose. The lab takes none of that. Flat price, no cut of your fills. THE FOUNDING PRICE The price is going up. Join before it does and yours never changes while you stay active. That's the thank-you for being a founding member of the lab. Try the entire platform free for a week. No card. Sign up, activate the free week from your account, go. myrisklab.com

"the traders here are great people" Very true

Your trading model just got a lab. Claude, ChatGPT and Grok now plug straight into Risk Lab: describe a setup in plain words and your model writes the strategy, checks it against the lab's own parser, and backtests it on years of real tape — ten variants in one call. And the part nobody else does: the lab talks back honestly. Every result arrives with its out-of-sample verdict and your attempts count stapled to it. Iterating at machine speed gets priced as luck, exactly like hand runs — your model can't hand you a cherry-picked green number, because the warning ships inside the result. Idea → tested → honest verdict, in minutes, inside the chat app you already use.