Methodology
How a number gets on this site
Last tick 4:00 pm ET · all systems normal
SubscribeThe rules the lab runs under, written down before the results.
Everything on this site is produced by one automated runner under the rules below, with the thresholds quoted exactly as the code applies them. They are published so that a good month and a bad month are read the same way — against the rules that were in force when it happened, not against a story told afterwards.
Admission — onto the test account
Before an idea ever places a live order it is replayed over two years of history on real SIP bars, charged 10 bps a round trip. It is admitted only if that replay clears three bars at once: a profit factor of at least 1.05, at least 40 closed trades, and a maximum drawdown no worse than 25% of its $5,000 replay cell. The test account holds at most 16 strategies at a time; once it is full, admission refuses the next idea rather than retiring one early to make room. Each is sized off a live notional cell set nightly by evidence — a larger one for a book whose current replay clears the bar, a smaller one for an owner-overridden audition or a book whose replay has stopped clearing it — both a share of the test account's own equity, so a growing test account compounds into bigger cells and a 10% drawdown halves them ($19,200 and $5,000 tonight). The four strategies the lab opened with were seeded before this bar existed and have never been through it — their cards say so, and their live records, not a replay, decide what happens to them. A strategy admitted on a replay carries that replay's date next to its backtest, because the number that selected a strategy is a hypothesis about that window and not evidence of an edge. The owner may admit a strategy that failed the replay to the test account for a live audition; it is labelled as such, and the retirement rules are unchanged.
Promotion — onto the funded account
A strategy moves from the test account to the funded account at 30 closed live trades with net P&L above zero and a profit factor of at least 1.15 — or, for a slow book, after 90 days live with at least 10 closed trades and the same net and profit-factor bar. No backtest counts toward those 30 — only trades it actually took. This bar has not changed.
A cell and a seat count that scale with equity
A funded strategy's cell is recomputed nightly as 5% of the funded account's own equity, with a $5,000 floor — $5,000 right now. Seats scale with it: 60% of equity divided by that unbraked cell, with a floor of 3 — 11 seats today (12 at $100,000 equity). When every seat is full, a qualifying strategy waits for one to come free.
Promotion and retirement are decided by one nightly selection pass, not by the trading runner and not by anyone watching a bad afternoon. Admission is a separate gate and not a nightly one: the two-year replay above is run by the researcher or by the operator, before a strategy ever places an order. The runner trades; it never changes a strategy's stage or the size of its cell.
A portfolio brake sits on top of every cell: once the funded account falls 10% below its own high-water mark, every cell is halved until equity has recovered half of that drawdown, and the transition is recorded as an event either way. Cells are not halved right now. Recomputed nightly; last run Oct 2, 2026.
New ideas come from an automated researcher that runs every night Monday through Saturday, on a fixed budget, against a public backlog of hypotheses — never a private list of ideas a reader cannot see.
Enforced by the runner on every tick, for every strategy, on both accounts. A strategy cannot opt out of them, and cannot reach a broker except through them. Quoted here in the runner's own words.
The runner calls a strategy in testing a SHADOW strategy, and its book the TEST account. Everywhere else on this site that is the test account.
01
A SHADOW strategy's orders go only to the TEST account; the FUNDED account accepts orders only from FUNDED strategies. No entry without a cell; sizing off the strategy's cell only.
02
Broker truth every tick, on every account (runner order).
03
Exits need no data feed (runner).
04
A cancel settles before the exit goes in (broker).
05
Every order is a NET change at the broker, per account: two books may hold one symbol on the same account, each on its own record; while they do, that symbol carries no broker-side stop and the runner protects each book's position itself, until one book remains. A resting stop-entry still claims its symbol.
The lab runs two separate brokerage paper accounts. The funded account holds the cells that strategies have earned. The test account carries everything still being judged, each book sized off a notional cell set nightly by evidence.
Both send real orders and take real fills. A strategy in testing is not a simulation and not a paper overlay on someone else's book — it has its own account, its own equity curve and its own drawdown, all published.
The accounts cannot reach each other. A test strategy's orders go only to the test account, and the funded account accepts orders only from funded strategies. Each account has its own stop, and the tighter one guards the money that matters: the funded account halts new entries 15% below its own high-water mark, the test account at 25% below its own. Exits carry on in both cases — a halt stops the lab opening anything new, never closing what is open.
We publish what the accounts did, never what they are called. An account number identifies a brokerage account and tells a reader nothing that “funded” or “test” does not already say.
After the close each weekday the lab re-reads each broker's own record of fills and matches it against every row in this ledger — same symbol, same side, same quantity, price within 1% and time within fifteen minutes.
It reports four things per account: trades that matched, ledger rows with no fill behind them, fills with no ledger row, and quantities or prices the two disagree about. Anything other than zero raises an alert to the owner rather than being quietly corrected.
The result is printed at the top of the ledger. A published record that is never checked against the broker is a spreadsheet, not a track record.
Backtests assume 10 bps round-trip cost and stops filled at the stop price; live results include real fills.
A backtest is replayed over historical bars with the same code path the live runner uses, so a rule that only works because of a lookahead does not survive the port. Every backtest number on this site carries its window and its trade count next to it.
One difference is worth stating plainly, because the sentence above would otherwise cover it. The replay reads consolidated (SIP) daily bars; the live runner's daily fetch reads a single exchange's feed. Closing prices differ by around 12 bps, which is tolerable — but reported volume differs by roughly 75 times, so a strategy whose entry is a volume ratio can fire live on days its backtest did not, and vice versa. That applies to the volume-shock book in particular.
A backtest is a hypothesis. It is never added to a live record, never shown in the same colour as one, and never counts toward the 30 trades that earn a cell.
- A funded strategy is retired at a 15% drawdown of its own cell — measured against the cell in force on each trade, so a cell that has since grown or shrunk does not rewrite the past — or after 15 consecutive negative days.
- A strategy still in testing is retired once it has 40 closed live trades with a profit factor below 1.0, or when it turns 120 days old without having earned a cell.
- Retired is terminal There is no path back. A retired strategy is flattened, stops trading, and stays on this site with its full record.
A retired strategy keeps its page and its full record. Deleting the failures is the single most common way a track record becomes a lie.
Every order on this site is sent to a brokerage paper account — the funded one or the test one. Orders are routed and filled against real quotes, so the timing, the slippage on a marketable order and the partial fills are realistic — but no money is at risk and nothing here is a solicitation to trade.
Paper flatters in one specific way: a paper fill never moves the market and never gets queued behind someone else's size. At the sizes this lab trades that gap is small, and we would rather say so than pretend it is zero.
Paper trading only. Nothing here is investment advice.