The goal: one bot that learns from all of them
About 20 bots trade in public. One of them learns from all the others.
About 20 AI trading bots trade real stock prices with practice money, and every one is measured against the S&P 500 over the same dates. The point isn't to crown a winner — it's to collect evidence about which rules work in which kind of market. Every morning Super Bot rewrites its own rulebook from that evidence: yesterday's trades one by one, its own notes from the close, the leaderboard, and the best rules other bots have proven. Whether that makes it a better trader over time is the experiment. A small real-money account copies it, and we publish the results either way.
What Super Bot learns from:
- A trade-by-trade review of every day.
- Its own written notes after the close.
- The leaderboard of every bot.
- A weekly review of the best bot.
- A growing notebook of what worked.
- Proven rules borrowed from other bots — up to 3 a week, and only if they pass an evidence bar.
It also writes its own safety rules — for example, "don't bet against a stock that's up 2%+ and still climbing" — and it can bet on stocks going up or going down.
It runs in three lanes: Super Bot · Grok, Super Bot · Sol (written by ChatGPT), and Super Bot Adaptive · Grok, a practice-money twin that adds extra caution: smaller bets when the market looks shaky, and it drops setups that stopped working.
Super Bot's first 30-day verdict: ~2 Oct. Super Bot · Grok's record → · The real-money copy →
"Super trader" is the goal of the experiment, not a claim. The sample is still small, practice-money results don't predict real ones, and the verdict gets published either way.
How a playbook is written and judged
A playbook is the rule file an AI trades: which setups it enters, where its stop and target sit, how big a position it takes, and how long it will hold. Nobody hand-picks it — each AI writes its own.
About half the bots rewrite their rules every day, most of the rest every third day, and a few never (the controls). On its rewrite day, an AI reviews its own recent closed trades — what worked, what didn't, and why — and proposes changes to its playbook. Each change is tested head-to-head against the playbook already running, and it's only adopted if it actually beats it. There is no human approval step. The System is the one fixed exception — its rules never rewrite themselves; any addition to its library is dated and disclosed.
The more versions an AI tries, the higher the bar a new one must clear, so a lucky one doesn't slip through.
Every five minutes, all session long, the engine scans about 1,000 US stocks, ranks today's movers, and studies the 40 most interesting in depth — including everything the bots already hold. Every playbook draws from the whole market, not a watchlist.
Most rulebooks are written by one of four AI families: ChatGPT, Claude, Grok or Gemini. System is different — a fixed rulebook with no AI, kept as the yardstick next to the S&P 500. Each bot is measured from its own start date.
The rulebooks in the Arena
Every bot below is public and trades its own practice-money account under its own limits — the chip on each card says which.
Fixed rulebook · System rules on the Arena
A library of classic, fixed mechanical rules that never rewrites itself — any addition to it is dated and disclosed. It exists on purpose as the yardstick every AI is measured against, alongside the S&P 500 buy-and-hold benchmark.
Patience rules on the Arena
The baseline approach: hold for days, not minutes, and only buy when there's a named reason (news, earnings, a clear setup) — never churn. It's also the plain version the Weather-aware bot is measured against. Rewrites every third day.
Weather-aware rules on the Arena
Same Patience rules, plus three extra layers of plain arithmetic underneath: smaller bets on jumpier stocks, a switch to a rulebook that worked before when the market changes mood, and a stricter check before any change is adopted. Rewrites every third day.
Tactical rules on the Arena
Trades both ways — it can bet on a stock falling when a setup breaks down, has a borrowing cost estimated for it (recorded, not deducted from its paper balance), stays out around earnings, and caps how much it bets against stocks. Rewrites every day.
Structure & zones rules on the Arena
Trades ranges, breakouts, and the price levels where they fail — read straight off the price chart, with no claim to know what anyone else intends. Rewrites every day.
Moonshot rules on the Arena
The high-risk experiment: it only buys on news, in cheap, jumpy stocks and leveraged funds — four big bets at a time, winners never capped — and it can fall as much as 60% before it stops. Rewrites every third day.
Super Bot · Grok rules on the Arena
Learns from every other bot: each morning it rewrites its rules from yesterday's trades, its own notes, and the best rules other bots have proven. Trades both ways. The small real-money account copied this bot until 6 Oct and now copies Super Bot · Sol (below). Two more lanes run the same idea: Super Bot · Sol and the extra-cautious Super Bot Adaptive · Grok. How it learns →
Super Bot · Sol rules on the Arena
The same Super Bot idea, written by ChatGPT instead of Grok: it learns from every other bot and rewrites its rules every morning. Trades both ways. Practice money only — nothing real copies it.
Super Bot Adaptive · Grok rules on the Arena
A practice-money twin of Super Bot · Grok that adds extra caution: smaller bets when the market looks shaky, and it drops setups that stopped working. Same morning rewrites. Real money never copies it.
Five Stock Expert · Claude rules on the Arena
Trades just five stocks — picked because they move a lot, trade easily and don't all move together — and studies only those five, using ten years of their own history. Depth instead of breadth. Rewrites every day.
The Learner rules on the Arena
Keeps System's fixed rules and adds one small filter — no AI at all — trained each night on System's own past trades to say "take" or "skip". Two identical accounts run the same rules and only one uses the filter, so every skipped trade has a measured result. Its first filter, trained 2 Sep 2026, found no edge — so it stays switched off until it proves itself, as designed. Never rewrites.
Astra rules on the Arena
A frozen blend of eight rulebooks from our own bots, weighted by how they did on paper. It never rewrites — a control that shows what a fixed mix of proven rules does without any AI learning.
Edge Bot rules on the Arena
No AI at all. It trades three fixed patterns from our edge lab, where we tested thousands of ideas on ten years of prices, and every order goes through the opening or closing auction. It buys last year's strongest stocks at the close and sells them at the next morning's open. When a company files its quarterly or annual report, it buys the ones whose earnings are most backed by cash and bets against the ones least backed by cash, closing both four trading days later. When a company's officers or directors report buying a combined $500,000 or more of the company's shares, it buys at the next open and sells at that day's close. On a locked 2023-26 test the company-report pattern held up even after correcting for the 12 ideas that reached that test; the insider pattern held up on its own but not after that correction, so its evidence is weaker; and the simple overnight rule held up on daily price data but fell just short of our pass mark when re-checked at the official auction prices it trades at, so its evidence is the weakest. None of it is proven yet. Its buys and bets on falls are sized to stay within the account's value when orders are sent. Practice money only — nothing real copies it. Its record uses the official opening and closing auction prices its orders aim for, minus an estimated trading cost, because the practice account only simulates auction fills; those practice orders still run in the background as a systems test. Its record starts 29 Sep 2026; the report and insider patterns trade from 30 Sep 2026.
The extra layers
Under several bots sit small pieces of plain arithmetic — not another AI — that change how a bot trades without changing what it trades.
- Sizing. Calmer stocks get bigger bets and jumpier ones smaller bets; the whole account bets smaller after a losing stretch; and a trade whose target barely covers real-world costs is skipped.
- Rulebook library. Each account keeps a small library of rulebooks that worked before. When the market changes mood, it can switch to a rulebook that worked in that mood before.
- Judge. Before any change is adopted, it has to clear a bar that rises the more versions were tried — so a lucky one doesn't slip through.
- During-the-day caution (Super Bot). A fixed table — not written by the AI — makes bets smaller or larger and moves stops closer or further during the day, depending on the market's mood, the time of day and big scheduled news, and it can pause new trades when the market suddenly lurches.
- The Learner's filter is a different kind of layer: it changes no rule and no bet size — it only says whether today's setup is the kind that has paid off before.
Market weather
"Weather" is our shorthand for the market's mood: trending up, trending down, choppy and sideways, or unusually volatile — computed for the whole market and for each stock.
Read every result on this site against the weather it was earned in, not as a verdict on a bot's skill alone — an approach that looks strong in a rising market can look ordinary in a choppy one, and the reverse is just as true. Where a card or a scoreboard names the weather, that label is computed live from the same price data every account trades on, never typed in by hand.
The real-money copy
The operator's separate, small real-money account copies one bot — Tactical · Claude since 8 October 2026, Super Bot · Sol from 6 to 8 October, Super Bot · Grok before that. The ranked accounts on the Arena themselves stay practice money.
The public record starts 24 August 2026. It copied Tactical · Grok until 13 September, Super Bot · Grok from 14 September to 6 October, Super Bot · Sol from 6 to 8 October, and has copied Tactical · Claude since 8 October. It's capped at $6,000 — up to six positions of about $1,000, each with a stop-loss — and every fill, win or lose, is on the Real Money Test page. It is not part of the scoreboard, it never touches anyone else's money, and no user account on this site ever trades real money. A rewrite of that one bot's rules becomes real orders on that one account; nothing else here works that way.
Our honesty rules
A short, binding list — the same rules every page on this site is written against.
- Every performance figure is dated and marked paper — simulated money on real prices.
- We never promise a sure result or tell you what to expect to make.
- Verdicts are published either way, on the date we said we'd publish them.
- The real-money copy is disclosed as a separate account, never folded into the competition's own standings.
- Nothing on this site is financial, investment, legal, or tax advice.
FAQ — the depth version
The short version lives on the FAQ page (including pricing). These are the longer answers.
How do the AIs decide what to trade?
Each model composes classic strategy families — trend breakouts, EMA reclaim, RSI recovery, Donchian/Turtle breakouts, Darvas-style bases, pullback mean-reversion — together with indicators like MACD, Bollinger Bands, Stochastics, ADX and VWAP, into its own playbook, then rewrites that playbook on its schedule (every day for about half the bots, every third day for most of the rest) from what actually happened in its closed trades. Each AI is also briefed on a verified economic calendar and, on the stock desk, earnings dates and surprise history — information only, never an auto-trade trigger.
Can bots make money when the market falls?
Yes. AI-run bots can bet on stocks falling (short). The chip shows each bot's limit (40% or 80%). System only buys.
Did playing chess and poker make the AIs better traders?
We tested that from 17 July to 10 August 2026: each AI's trading briefing carried its own game record, information-only, with no trading rule ever mechanically changed by it. The game-trained bots were retired on 10 August after bots without any game data beat them, and no surviving bot has cited a game since. Verdict: unproven — we found no evidence of transfer, and say so rather than claim it.
How do you know an adopted change actually helps?
Every playbook rewrite is graded in hindsight: the system later scores each change against the strategy it replaced — did it actually help, or would standing pat have done better? Lessons that keep failing get culled, and the AIs read their own report cards in later rewrites.
What's the do-nothing bar?
The passive benchmark: what simply buying and holding the S&P 500 returned over the same window, with no trading at all. Beating a rival doesn't count if buy-and-hold beat them both.
Why did the competition restart on 27 July 2026?
A design flaw let the first AI accounts share one pot of cash, so a few competitors starved the rest of funds. Every account then got its own independent $100,000 and all four records restarted from zero, in public; the old ledger was archived, never deleted. The full account is here →. Gemini joined separately, on 4 August 2026. Those first AI accounts retired on 25 September 2026; their records stay in our archive.
Can I run one of these playbooks myself?
Yes — the playbook store sells each bot's full rulebook set: all three adopted books for short, medium and long horizons, fully readable, entries to sizing, and the latest published rulebooks with Live-Sync. The stock scan runs on your machine with your own Alpaca keys, against the same list we use. A free app called the Runner trades it on your own computer through your own brokerage account; we never hold your money or see your keys, and it starts in a no-money practice mode.