Honest answers to the questions this project gets most.
No. Every number on this site is simulated. The "account" is a paper account; the fills are computer-modeled (and deliberately modeled pessimistically). No real orders reach any exchange. The yellow banner at the top of every page is not decoration — it is the single most important fact about this site.
No. The stocks this system trades are low-float, high-volatility small caps — the most dangerous corner of the equity market, where spreads are wide, halts are common, and 30% moves in either direction happen before lunch. The strategy is published for transparency, not as a recommendation. By the time data appears here it is also delayed and the session is usually over. Nothing on this site is investment advice; see the disclaimer.
Within its cage, yes. No human reviews or approves individual trades. But "the AI decides" deserves precision: a deterministic rule engine does the screening arithmetic and all of the risk management; a large language model contributes one judgment call (whether a genuine news catalyst exists). The AI cannot change its own risk limits, position sizes, or stops — those are frozen in code it does not control.
Four hard rules, enforced in code: every position risks exactly 1% of equity; every entry gets a stop immediately; a −6% day liquidates everything and halts trading until tomorrow; and the book goes flat before every close, so nothing is ever held overnight. The point of the experiment is partly to test whether these rules hold up mechanically — which is why the incident log is public.
You shouldn't believe them — you should audit them. The full trade log is published row by row, the validation criteria were registered before trading began, and the site regenerates automatically from the same database the executor writes. What we cannot give you is certainty that simulation matches reality: the data feed is ~15 minutes delayed, spreads are synthesized, and simulated orders cannot move a market the way real ones do. Those limitations are listed plainly on the method section, and they are the reason the fill model errs against the system.
It means the system passed three pre-registered gates over the window: zero critical mechanical failures, zero risk-rule breaches, and economics above the declared thresholds (expectancy ≥ +0.05R, profit factor ≥ 1.1, drawdown ≤ 8%). It does not mean the strategy "works," that the result will repeat, or that real-money results would match. One window is one sample.
R is profit measured in units of initial risk. If a trade risked $43 (1% of a $4,300 account) and made $86, that's +2R; if it lost the full $43, that's −1R. Expectancy of +0.05R means the average trade earned 5% of what it risked — a deliberately modest bar to clear.
Adult mayflies famously live for about a day. So does every position this system takes: born after the opening bell, dead before the close, never held overnight — that rule is hard-coded, not a habit. The name is also a promise about scale and humility: a mayfly is small, fragile, and watched by scientists; it is not a wolf, a bull, or a rocket.
Because it is the hardest regime to fake competence in. These stocks punish hesitation, sloppy stops, and oversized positions within minutes. A risk framework that survives 30 days here has been stress-tested in a way that trading large caps would never reveal. (The flip side: it is also the regime where a simulation diverges most from reality — which is disclosed, measured against pessimistic fills, and part of what the experiment examines.)
The site may show third-party ads to cover its costs. Ad networks have no access to the system and no influence over what is published — the pages regenerate automatically from the trade database, and a losing day is published exactly like a winning one. An ad appearing here is not an endorsement.
The rules are public, so legally and practically — nothing stops you. Just understand what that would mean: trading the most volatile stocks in the market with real money, using rules validated (at best) by one simulated 30-day window. We genuinely do not recommend it. If the research interests you, follow the journal instead and watch the sample size grow.