158 days from the first line of code to V17.
600–900 hours of work275 experiments on recordScroll to continue ↓
01 / The idea
We set out to build an algorithmic system that trades by itself. It watches the market, makes its own decisions, places and manages its own orders, and reports what it did. No one sits at the screen.
Building it once was never the goal. We always improve it. Each new idea is tested against the current version on the same data, and goes live only when it earns its place. Every version is saved, so any step can be undone.
02 / How it behaves
This page does not describe what the bot looks for in the market. It describes how the bot conducts itself, every day, with no one watching.
It runs in the cloud without anyone at a screen. After a fault it restarts itself and picks up its saved state.
250 trades in the 2026 simulation, about 25 a month. The rest of the time it holds nothing.
It never stacks trades on top of each other.
Every position has a stop order resting at the broker, so the exit exists even if the bot goes down.
When the daily loss limit is reached, it takes no new trades for the rest of the session.
Positions are closed before the market's daily close and before the weekend, every time.
It stays out around high-impact economic releases and closes any open position before they arrive.
If its data is missing or stale it takes no new trades, while any open position stays managed.
An email for each trade and one for each day, including days with no trades. A silent day counts as an alarm.
Results are reconciled to the broker's balance. The bot's own estimates are never treated as the truth.
03 / 2026, month by month
What V17 would have made on one $150,000 funded account at a prop firm, opened on 1 January 2026, with nothing withdrawn. Fees and slippage are included. The figures run to 30 October, the last trading day of the month.
Why a prop firm. We wanted to test the bot with real orders without risking more of our own money, so we went with a proprietary trading firm. The $150,000 is the firm's account, not our money: we pay a fee for it and trade under the firm's strict limits on daily loss and total drawdown. If the bot breaks a limit, the account ends and what we lose is the fee. If it passes the firm's evaluation, it trades a funded account and we keep 90% of the profit.
| Month | V17, simulated | Running total |
|---|---|---|
| January | +$9,874 | +$9,874 |
| February | +$8,586 | +$18,461 |
| March | +$57,241 | +$75,702 |
| April | +$34,733 | +$110,435 |
| May | +$66,054 | +$176,489 |
| June | +$108,295 | +$284,784 |
| July | +$80,479 | +$365,264 |
| August | +$32,799 | +$398,063 |
| September | +$11,858 | +$409,920 |
| October | +$19,356 | +$429,277 |
| Total | +$429,277 |
04 / October 2026, day by day
Every trading day of October on the same $150,000 funded prop-firm account, simulated with V17's rules: 29 trades. A day runs from 6 pm to 6 pm New York, the way the broker counts it.
| Day | Result | October so far |
|---|---|---|
| Thu 1 | +$6,503 | +$6,503 |
| Fri 2 | −$3,694 | +$2,809 |
| Mon 5 | −$2,617 | +$192 |
| Tue 6 | +$3,407 | +$3,599 |
| Wed 7 | flat | +$3,599 |
| Thu 8 | flat | +$3,599 |
| Fri 9 | +$23 | +$3,622 |
| Mon 12 | −$452 | +$3,170 |
| Tue 13 | +$864 | +$4,035 |
| Wed 14 | flat | +$4,035 |
| Thu 15 | +$7,927 | +$11,962 |
| Fri 16 | +$11,490 | +$23,452 |
| Mon 19 | −$4,319 | +$19,133 |
| Tue 20 | +$2,474 | +$21,607 |
| Wed 21 | −$1,551 | +$20,056 |
| Thu 22 | −$2,836 | +$17,220 |
| Fri 23 | +$2,981 | +$20,201 |
| Mon 26 | flat | +$20,201 |
| Tue 27 | −$7,644 | +$12,557 |
| Wed 28 | flat | +$12,557 |
| Thu 29 | −$4,516 | +$8,040 |
| Fri 30 | +$11,316 | +$19,356 |
| Total | +$19,356 |
29 trades: 13 won, 16 lost. Money won +$51,813, money lost −$32,457.
| Day | Result | October so far | Contracts | How it ended |
|---|---|---|---|---|
| Thu 1 Oct | +$2,330 | +$2,330 | 14 | Target |
| Thu 1 Oct | +$4,173 | +$6,503 | 50 | Target |
| Fri 2 Oct | −$3,449 | +$3,055 | 61 | Stop |
| Fri 2 Oct | −$29 | +$3,026 | 14 | Closed on schedule |
| Fri 2 Oct | −$217 | +$2,809 | 24 | Closed on schedule |
| Mon 5 Oct | −$2,617 | +$192 | 24 | Stop |
| Tue 6 Oct | +$3,407 | +$3,599 | 24 | Closed on schedule |
| Fri 9 Oct | +$23 | +$3,622 | 24 | Time limit |
| Mon 12 Oct | −$452 | +$3,170 | 7 | Time limit |
| Tue 13 Oct | +$4,899 | +$8,069 | 58 | Target |
| Tue 13 Oct | −$4,034 | +$4,035 | 37 | Stop |
| Thu 15 Oct | +$1,886 | +$5,921 | 25 | Target |
| Thu 15 Oct | +$6,040 | +$11,962 | 38 | Target |
| Fri 16 Oct | +$3,468 | +$15,430 | 25 | Target |
| Fri 16 Oct | +$8,022 | +$23,452 | 39 | Target |
| Mon 19 Oct | −$728 | +$22,724 | 7 | Stop |
| Mon 19 Oct | −$3,591 | +$19,133 | 69 | Stop |
| Tue 20 Oct | +$2,474 | +$21,607 | 25 | Target |
| Wed 21 Oct | −$1,551 | +$20,056 | 25 | Stop |
| Thu 22 Oct | −$2,705 | +$17,351 | 26 | Stop |
| Thu 22 Oct | −$131 | +$17,220 | 39 | Stop |
| Fri 23 Oct | −$632 | +$16,588 | 26 | Stop |
| Fri 23 Oct | +$3,613 | +$20,201 | 26 | Closed on schedule |
| Tue 27 Oct | −$3,653 | +$16,548 | 73 | Stop |
| Tue 27 Oct | −$3,992 | +$12,557 | 38 | Stop |
| Thu 29 Oct | −$4,516 | +$8,040 | 95 | Stop |
| Fri 30 Oct | +$3,918 | +$11,959 | 26 | Target |
| Fri 30 Oct | −$161 | +$11,798 | 25 | Stop |
| Fri 30 Oct | +$7,558 | +$19,356 | 40 | Target |
| Total, 29 trades | +$19,356 |
Swipe the table sideways to see every column.
05 / V17 against V15
The same data, the same account, the same costs. V15 went live on 26 October and V17 two days later. Of the 17 ideas chosen for V17, 13 are live now; the other four are built and switched off, or not built yet.
| 1 Jan to 30 Oct | V15 | V17 |
|---|---|---|
| Simulated profit | +$308,397 | +$429,277 |
| Trades | 178 | 250 |
| Winning trades | 54.5% | 55.2% |
| Profit factor | 2.90 | 3.11 |
| Profit per trade | +$1,733 | +$1,717 |
| Largest drawdown | $15,043 | $15,411 |
| Worst day | −$8,998 | −$8,998 |
| Best day | +$29,179 | +$32,781 |
| Days up / down | 77 / 55 | 100 / 51 |
| Losing months | 1 of 10 | 0 of 10 |
Now the same bot with ordinary money. No prop firm, no firm's rules, just $10,000.
06 / A $10,000 account
An example with ordinary money at a regular futures broker: $10,000 on 1 January 2026, nothing withdrawn, no prop-firm rules. Each account risks the same share of its balance on every trade, so the trade size grows with the balance and shrinks when it falls. If the money runs out, the account is over. The bot's own habits stay: one position at a time, a daily stop at twice the trade's risk, flat before the daily close.
The safest line and the slowest. Its largest fall was 5.3%, so most of the room it could safely use goes unused.
Far more money, with a much deeper fall: 24.1% from its high and −10.1% on its worst day. With the same days in 9,000 other orders, 3.1% fell by a third at some point.
Only 17% more money than 5%, because both reach the 100-contract limit, with deeper falls. With the same days in 9,000 other orders, 44% fell by a third at some point and 5.0% fell by half.
The best spot
1% is too little. 5% earns more but falls much further. 10% adds almost nothing. We also ran 2% and 3% to find the spot between them.
At 3% the account ended at $556,414 with a largest fall of 14.6%. It is the highest level where none of the 9,000 other orders of the same days fell by a third, against 3.1% at 5% and 44% at 10%.
The scale multiplies instead of adding, so a 10% fall looks the same size anywhere on the chart. The higher lines bend over when they reach the 100-contract limit: from there they grow by a steady amount, not a steady share.
| Month | Start | Trades | Won / lost | Money won | Money lost | End | Change | Largest size |
|---|
Swipe the table sideways to see every column.
| Risk on every trade | 1% | 3% | 5% | 10% |
|---|---|---|---|---|
| V17, as simulated | ||||
| Balance on 30 October | $39,155 | $556,414 | $846,197 | $987,788 |
| Money won | +$48,347 | +$920,526 | +$1,323,478 | +$1,590,062 |
| Money lost | −$19,192 | −$374,112 | −$487,281 | −$612,274 |
| Largest fall from a high | 5.3% | 14.6% | 24.1% | 31.3% |
| Worst day | −2.1% | −5.8% | −10.1% | −12.9% |
| Worst single trade, share of the balance | −1.1% | −3.5% | −5.5% | −11.1% |
| Risk actually taken, typical trade | 0.8% | 2.7% | 2.5% | 1.7% |
| Trades at the 100-contract limit | 0 of 258 | 108 of 260 | 156 of 260 | 217 of 260 |
| V17, the same days in 9,000 other orders | ||||
| Ending balance, 9 orders in 10 | $26.9k to $65.3k | $271k to $829k | $563k to $1.10M | $740k to $1.27M |
| Orders that fell by a third at some point | none | none | 3.1% | 44% |
| Orders that fell by half at some point | none | none | none | 5.0% |
| V15, the same risk | ||||
| Balance on 30 October | $25,683 | $200,872 | $465,158 | $618,785 |
| Largest fall from a high | 5.3% | 15.3% | 19.1% | 30.5% |
| Worst day | −2.0% | −6.0% | −10.6% | −20.0% |
Swipe the table sideways to see every column.
Trade size stops at 100 micro contracts, the most the bot is configured to trade. The 3% account reached it on 18 June, the 5% account on 30 March and the 10% account on 24 February. After that their real risk keeps falling as the balance grows, so a typical trade risked 2.7%, 2.5% and 1.7%.
These sizes assume the broker asks about $300 a contract or less while the position is open. Published day-trading rates for this contract run from about $50 to $300; they vary by broker and by hour, some brokers offer none, and we have not confirmed one. At the full overnight margin, up to about $3,800 a contract in the figures we found, $10,000 holds two contracts: a 5% setting then really risks about 2% and ended at $137,136.
Trading costs are charged: about $46,180 on the 3% account. Taxes, market-data fees and server costs are not. The bot has never traded an account like this one: it is an example, not a product we run.
07 / How we got here
No data, no model and no track record. We began with the little data we could get, tested it, searched the internet for what others had learned, and kept only what survived the tests.
The first line of code. No market data and no tested idea.
The first price data, bought from a market-data vendor. The first tests ran on about 30 days of it.
The first full test across six months. Results that looked strong on 30 days did not all survive.
The first live trade with real orders on a prop-firm evaluation account, so the bot could be tested without risking more of our own money.
History extended back to March 2024, so ideas could be judged in more than one kind of market.
We searched outward: academic papers, public market data, trading videos, AI research tools and a nightly research agent. Every idea went into a written ledger, and most of them failed.
We audited our own simulator and found its results overstated. Correcting it cut the headline figure by 38%. All figures on this page use the corrected simulator.
V15 went live, and about 370 ideas went back through the full simulator, on a desktop PC and 12 cloud machines at once.
V17 went live.
08 / The horizon
The lines stop where the data stops, and we draw no projection past that point. The next two marks are forward tests: months the rules have never seen, run in the same simulator and set beside the broker's real balance.
The current account is still in the prop firm's evaluation. It has to reach the firm's profit target while staying within its drawdown limit.
New data that no rule was chosen on: V17 against V15, on the new months only.
A longer stretch of new data, with the real broker balance beside the simulation.
A funded account held for income, with the same loop still running: new ideas tested, the survivors shipped, the results checked at the broker.
