> ## Content Index > Fetch the complete content index at: https://thomasadair.ghost.io/llms.txt > Use this file to discover other available public pages before exploring further. # You Only Get One Draw From the Market - URL: https://thomasadair.ghost.io/you-only-get-one-draw-from-the-market/ - Published: 2026-09-11T17:17:05.000Z - Updated: 2026-09-11T17:17:05.000Z - Author: Thomas Arthur Adair - Tags: short-form-note, #short-form-note *Futures Friday — Thomas Adair* You can spend a month making one strategy better, or you can spend it building a second strategy that fails at a different time. Those are different bets, and only one of them survives being wrong. I worked through this once already, on prompts. The trade came out as: optimizing buys you a better single prompt for the price of building a metric, and diversifying buys you a variance estimate for the price of extra tokens and no build. For a solo builder the second usually wins, because it’s cheap enough to actually happen. The same trade shows up one level up, on strategies. The odds are worse there, and the stakes are real money. ## Different instruments is not diversification I’ve been reading Angus — he writes *Algo Trading Success*, he won a Quantiacs championship and the allocation that came with it, and he’s further along the public-teaching path than I am. He’s in my lane through a different door: equities and QQQ, retail-accessible, while I’m in futures and order flow. He makes an argument I think is correct and underweighted. Build strategies with low correlation to *each other*, and measure that correlation between the equity curves — the running plot of each strategy’s account value over time — not between the underlying assets. Those are different measurements and they give different answers. Trade the same logic across ten symbols and you own ten copies of one bet. Your positions look diversified. Your P&L isn’t. When the condition your logic assumes goes away — the volatility regime shifts, the session character changes, the flow that made your entries good dries up — all ten go quiet together. The assets were uncorrelated. The *edges* were the same edge. Two strategies are genuinely diversified when they break for unrelated reasons. A mean-reversion book and a trend book on one instrument are more diversified than one mean-reversion book across a whole exchange, because their bad months come from opposite conditions. So the question isn’t “what else can I trade.” It’s what would have to be true for this to stop working, and whether I own anything that’s fine when that happens. ## Why the odds are worse here than on prompts With a prompt, optimizing is expensive but possible. You can define a metric, run the thing a thousand times, and get a real answer. The reason I mostly don’t is cost, not impossibility. A strategy doesn’t give you that option at all. The scoreboard is the market. You get one draw from the distribution, you can’t rerun the year, and the thing you’d be tuning is the thing keeping score. There is no cheap dev set for 2026\. So the argument that was a cost trade-off on prompts becomes something closer to a structural fact on strategies: you will never accumulate the sample size to optimize your way to confidence, and the only real protection is owning something uncorrelated for when you’re wrong. Same trade. Worse odds. Same answer. ## The part I’m adopting from him Angus publishes monthly recaps of real-money results. Not backtests. What actually happened. He also runs a Monte Carlo validation battery over his strategies — repeatedly resampling the trade sequence to see how much of the result was the edge and how much was ordering — and he reports the failures. One of his published strategies fails his own statistical-edge test and passes his regime-independence test, and he says so, in public, with the strategy still on the list. That’s the practice I’m adopting. Not his catalogue and not his format for selling it — we’re selling different things, and he’s built a real business handing people implementations while I’d rather hand people the thinking. But putting real numbers where readers can see them, including the ones that didn’t work, is what earns trust, and almost nobody does it. So the Digest gets a live-results section. Real money, monthly, losing months included. ## What it costs Diversification isn’t free and I won’t sell it as free. A second strategy is a second thing to maintain, a second set of gates, a second way to be wrong when you’re not watching. Two mediocre strategies can lose more than one good one. The argument was never “more is better” — it’s that returns add up while failure modes concentrate, and most people optimize until they own exactly one. So watch the correlation of the curves. If a new strategy’s curve tracks the old one, you didn’t diversify… you built the same bet twice and doubled the maintenance. The number that matters isn’t how good the best month looks. It’s how many different ways the book survives a bad one. ## Sources - Thomas Adair, *Diversify or Optimize Your Prompts?* — the prompt-level version of this trade, and where the cost argument comes from. - Angus (@angusslq), *Algo Trading Success* — https://angusslq.substack.com — Quantiacs Q22 championship and allocation; monthly real-money “Quant Live Trading Systems” recaps; Monte Carlo validation series. - Angus, *Warren Buffett’s “Don’t Put All Eggs In One Basket.” You Think You Get It. You Don’t.* — https://angusslq.substack.com/p/warren-buffetts-dont-put-all-eggs — the real-diversification-is-correlation-under-stress argument. (Full framework is paid; the public preview states the core point paraphrased above. Nothing quoted.)