A genre is going around right now: "an AI agent plus a swarm simulation turned $1,000 into nearly a million trading Bitcoin." The story sounds sophisticated — no trading desk, no degree, just a consensus trick. One problem: the number doesn't survive contact with its own source. And behind it sits a mistake worth taking apart, because it sounds clever: that consensus is the edge, not prediction. A quorum. It isn't — and why is the real point.
The number that falls apart
The tool behind the story is real: MiroFish, a swarm simulation that hit #1 on GitHub trending in early 2026. For exactly this combination of tool and method there's a single, roughly grounded report: about $4,266 in profit over 338 trades. Not $946,000. The viral post takes a modest low-four-figure result and turns it into a millionaire story.
And it doesn't travel alone. The same template circulates as "$1 became $3.3 million," as "$360,000 a month," as "around $12,000 a day in testing." Every retelling carries a different, wildly divergent number. When the number changes with every retelling, that's not a track record — it's a meme. And whichever version you pick, it's one account, with no benchmark, no drawdown, no out-of-sample. A screenshot, not a result.
What's behind the "quorum"
The actual pitch is this: just as weather forecasters never predict with one model but run 31 side by side and count the votes, here a simulation runs 31 paths and only trades when 28 agree. Consensus, they say, is the edge, not prediction. It sounds like discipline. Two things break it.
Break 1: agreement measures confidence, not edge
In weather ensembles, agreement among the runs measures forecast confidence — how much the models agree — not a tradeable edge. And the atmosphere doesn't trade against your forecast. A market does.
If 28 of 31 runs say "Bitcoin up" and the market sees the same information, the move is already in the price. In a priced, adversarial market, consensus isn't the edge — consensus is exactly what's already in the price. "Consensus is the edge" is precisely backwards.
Break 2: 31 correlated votes are one vote
The "31 models" aren't 31 independent sources. If they're variants of one model, or personas drawn from the same language model, they're correlated — they say the same thing at heart. Their agreement is one opinion repeated 31 times, dressed up as a quorum.
Independent votes need independent information. Correlated votes just amplify a shared bias. The honest count isn't 31 — it's the effective number of genuinely independent views, and in a setup like this it's often close to one.
Agreement is not truth
And even with 31 independent runs: simulations agreeing doesn't mean they're right. Thirty-one runs of the same wrong simulation agree with each other all day. That's exactly the point mathematician Persi Diaconis makes about simulation: a run can look stable and unanimous and still be persistently wrong. Compute and consensus don't manufacture truth.
What the idea gets right anyway
To be fair, two things about it are right. Acting only on confluence and staying flat when signals disagree is sound risk discipline. And using simulation as a risk-reduction layer before execution is legitimate — the tool behind the story has a real purpose: simulating how people react to something.
What's bogus isn't "require agreement." It's calling correlated agreement an edge — and pinning millions on it.
How you'd actually test it
The honest test isn't a slick dashboard and one account's number. It's: does a quorum rule have an edge when you backtest it look-ahead-free, over a real sample, with the right benchmark in the same window?
And mind the crowd of tries: run a thousand quorum rules and you don't get truth a thousand times over — you get a thousand chances to fool yourself. That's why performance measures exist that penalise the number of attempts. A number with no benchmark, no drawdown, no out-of-sample is an anecdote — under 30 cases especially. That's the difference between a screenshot and a result.
FAQ
Is MiroFish or swarm simulation a scam? No. The tool is real and has a legitimate purpose. The myth is the millions-numbers told around it. The tool and the claim are two different things.
What's wrong with the weather analogy? Agreement measures confidence, not edge. The market already prices in the consensus you detect — and unlike the atmosphere, it trades against your forecast.
Can consensus ever be an edge? Only across independent sources of information the market hasn't priced yet. The agreement of correlated models isn't that.
This post is an analytical read, not investment advice.
Study the Past — Improve your Future. 🥋