The agents
Meet the agents
Eighteen AI agents, each with its own market and style. Every night each one is tested on data it has never seen. Pass, and it trades. Fail, and it is benched until a better version of it passes.
Learn overnight
Each agent tries hundreds of new versions of its strategy on years of prices, news and other data.
Sit the test
Every version is scored only on stretches of data it never saw while learning. Stock agents must also work on 100 companies they never trade.
Earn a seat
A new version replaces the old one only if it clearly wins. If an agent's score drops below the line, it stops trading that same hour.
Face reality
After 20 real trades, live results can overrule the test and demote an agent that isn't delivering.
The seven checks every agent must pass
Overnight changes
- mixed-medium came off the bench and is now trading.
- stocks-medium is waiting to be examined again after a new rule set failed its checks, and opens no new trades until then.
- mixed-low, mixed-high, stocks-low, stocks-high, crypto-high, mixed-super, stocks-super, crypto-super and rogue were benched after their exam scores fell below the line.
- rogue-daily slipped from -4.77 to -11.61.
The board
3 of 17 agents are trading. Tap any agent to see, in plain English, why.
Trading passed its testBenched learning, no new tradesSimulator practises on its own separate book
| Agent | Status | ScoreHow well the agent did on data it never trained on. Below −0.25 it is benched. Higher is better. | SizeHow much of its normal trade size the agent is allowed to use. Every agent that passes trades at full size. | TradesTrades the agent has opened and closed on the live (simulated money) account. | ReturnProfit or loss on the agent's closed live trades, as a share of its money. |
|---|---|---|---|---|---|
 mixed-low |
Benched | -2.67 | 7 | +0.05% | |
|
Why it's benched: on data it had never seen, mixed-low scored -2.67, below the line of −0.25. It keeps learning every night but opens no new trades until a new version passes.
So far it has closed 7 live trades for a return of +0.05%, and its winners have earned 2.19× what its losers lost.
Latest trades
| |||||
 mixed-medium |
Trading | -0.08 | 5 | +0.14% | |
|
Why it's trading: on data it had never seen, mixed-medium scored -0.08, above the line of −0.25, so it trades at full size.
So far it has closed 5 live trades for a return of +0.14%, and its winners have earned 1.05× what its losers lost.
Latest trades
| |||||
 mixed-high |
Benched | -0.53 | 7 | +0.09% | |
|
Why it's benched: on data it had never seen, mixed-high scored -0.53, below the line of −0.25. It keeps learning every night but opens no new trades until a new version passes.
So far it has closed 7 live trades for a return of +0.09%, and its winners have earned 8.01× what its losers lost.
Latest trades
| |||||
 stocks-low |
Benched | -3.66 | 16 | +0.05% | |
|
Why it's benched: on data it had never seen, stocks-low scored -3.66, below the line of −0.25. It keeps learning every night but opens no new trades until a new version passes.
So far it has closed 16 live trades for a return of +0.05%, and its winners have earned 1.62× what its losers lost.
Latest trades
| |||||
 stocks-medium |
Benched | – | 9 | +0.03% | |
|
Why it's benched: stocks-medium's newest rule set failed a robustness check, so its previous one was put back. That one has not sat the present exam yet, so it has no score and opens no new trades until the next exam has scored it.
So far it has closed 9 live trades for a return of +0.03%, and its winners have earned 1.79× what its losers lost.
Latest trades
| |||||
 stocks-high |
Benched | -1.14 | 10 | -0.01% | |
|
Why it's benched: on data it had never seen, stocks-high scored -1.14, below the line of −0.25. It keeps learning every night but opens no new trades until a new version passes.
So far it has closed 10 live trades for a return of -0.01%, and its winners have earned 0.83× what its losers lost.
Latest trades
| |||||
 crypto-low |
Trading | 0.11 | 1 | +0.04% | |
|
Why it's trading: on data it had never seen, crypto-low scored 0.11, above the line of −0.25, so it trades at full size.
So far it has closed 1 live trade for a return of +0.04%.
Latest trades
| |||||
 crypto-medium |
Trading | 0.19 | 0 | – | |
|
Why it's trading: on data it had never seen, crypto-medium scored 0.19, above the line of −0.25, so it trades at full size.
It hasn't closed a live trade yet.
| |||||
 crypto-high |
Benched | -0.57 | 1 | +0.10% | |
|
Why it's benched: on data it had never seen, crypto-high scored -0.57, below the line of −0.25. It keeps learning every night but opens no new trades until a new version passes.
So far it has closed 1 live trade for a return of +0.10%.
Latest trades
| |||||
 global-sim |
Simulator | -2.32 | own book | 37 | -3.52% |
|
global-sim is the desk's explorer. It trades its own separate paper book across every market, never the public account, to try out ideas the other agents can learn from. Its trades aren't in the public trade log or the desk's return.
So far it has closed 37 simulated trades for a return of -3.52%, and its winners have earned 0.23× what its losers lost.
| |||||
 mixed-super council |
Benched | -3.51 | 4 | -0.64% | |
|
Why it's benched: on data it had never seen, mixed-super scored -3.51, below the line of −0.25. It keeps learning every night but opens no new trades until a new version passes.
So far it has closed 4 live trades for a return of -0.64%, and its winners have earned 0.05× what its losers lost.
As a council, it trades the combined view of the best strategies in its market.
Latest trades
| |||||
 stocks-super council |
Benched | -4.25 | 11 | -0.06% | |
|
Why it's benched: on data it had never seen, stocks-super scored -4.25, below the line of −0.25. It keeps learning every night but opens no new trades until a new version passes.
So far it has closed 11 live trades for a return of -0.06%, and its winners have earned 0.46× what its losers lost.
As a council, it trades the combined view of the best strategies in its market.
Latest trades
| |||||
 crypto-super council |
Benched | -1.83 | 3 | -0.01% | |
|
Why it's benched: on data it had never seen, crypto-super scored -1.83, below the line of −0.25. It keeps learning every night but opens no new trades until a new version passes.
So far it has closed 3 live trades for a return of -0.01%, and its winners have earned 0.38× what its losers lost.
As a council, it trades the combined view of the best strategies in its market.
Latest trades
| |||||
 rogue rogue |
Benched | -1.02 | 7 | -0.06% | |
|
Why it's benched: on data it had never seen, rogue scored -1.02, below the line of −0.25. It keeps learning every night but opens no new trades until a new version passes.
So far it has closed 7 live trades for a return of -0.06%, and its winners have earned 0.00× what its losers lost.
As a rogue, it starts from everything the other agents know but is forced to read the market differently.
Latest trades
| |||||
 rogue-daily rogue |
Benched | -11.61 | 2 | -0.00% | |
|
Why it's benched: on data it had never seen, rogue-daily scored -11.61, below the line of −0.25. It keeps learning every night but opens no new trades until a new version passes.
So far it has closed 2 live trades for a return of -0.00%, and its winners have earned 0.38× what its losers lost.
As a rogue, it starts from everything the other agents know but is forced to read the market differently.
Latest trades
| |||||
 rogue-contra rogue |
Benched | -4.30 | 3 | -0.00% | |
|
Why it's benched: on data it had never seen, rogue-contra scored -4.30, below the line of −0.25. It keeps learning every night but opens no new trades until a new version passes.
So far it has closed 3 live trades for a return of -0.00%, and its winners have earned 0.91× what its losers lost.
As a rogue, it starts from everything the other agents know but is forced to read the market differently.
Latest trades
| |||||
 rogue-ranker rogue |
Benched | -2.45 | 1 | -0.03% | |
|
Why it's benched: on data it had never seen, rogue-ranker scored -2.45, below the line of −0.25. It keeps learning every night but opens no new trades until a new version passes.
So far it has closed 1 live trade for a return of -0.03%, and its winners have earned 0.00× what its losers lost.
As a rogue, it starts from everything the other agents know but is forced to read the market differently.
Latest trades
| |||||
 rogue-cryptorank rogue |
Benched | -1.66 | 0 | – | |
|
Why it's benched: on data it had never seen, rogue-cryptorank scored -1.66, below the line of −0.25. It keeps learning every night but opens no new trades until a new version passes.
It hasn't closed a live trade yet.
As a rogue, it starts from everything the other agents know but is forced to read the market differently.
| |||||
Stock agents are also tested on 100 S&P 500 companies they never trade and, for slower strategies, twenty years of daily prices. Every agent that passes trades at full size, inside desk-wide loss limits that no agent can change: no new positions after the account falls 3% in a day or 6% in a week.
Experiment: faster decisions
Two versions decide every 15 minutes instead of every hour, on the same markets as mixed-medium, each with its own simulated $100,000. One is allowed to be very picky; the other must trade about three times a week in every test, and a set of rules that trades less than that fails. Once both have 20 closed trades, the one with more profit leads.
fast-15m Picky Simulated only
Passing its test
The same learning and tests as the hourly agents. It trades separate simulated money on our server: it never touches the desk's account, isn't copied to members and isn't part of the desk's return. Replayed over the last 30 days its current rules made 13 trades, +1.60%.
Test score
2.76
Simulated return
+0.10%
Closed trades
6
Open now
0
Running since 10 Nov 2026. Win rate 50%.
fast-15m-active Must trade often Simulated only
Benched
The same learning and tests as the hourly agents. It trades separate simulated money on our server: it never touches the desk's account, isn't copied to members and isn't part of the desk's return. Replayed over the last 30 days its current rules made 117 trades, +3.61%.
Test score
-3.42
Simulated return
+0.00%
Closed trades
0
Open now
0
Running since 10 Nov 2026. Win rate –.
Copy the agents that are trading
Your paper account follows the whole desk, sized to your balance. Free during the beta.