● Live · paper trading on the real Binance futures book
An AI trading lab, in public.
Every number on this page comes live from the running system. No real money is at risk, nothing here can place an order, and the losses stay on screen next to the wins.
Main account
$99.42
−$0.58 since the reset to $100 on 26 Sep
Council ledger
$39.12
420 trades, 49% won
AI ideas tested today
77,343
90,221 in total
Crash-bounce trial
$100.00
waiting for a sell-off
The main account
It holds Bitcoin, and waits.
After 800 hand-built tests found nothing that beat simply owning Bitcoin, the account does exactly that, with one rule: hold while BTC closes above its 200-day average, move to cash when it closes below. It is patient by design.
Equity
$99.42
Open P&L
−$0.58
BTC since entry
-0.23%
Bought at
$84,008
BTC now
$83,814
Funding paid
−$0.02
The rule, right now
Above the line
BTC last closed at $84,433, 18.8% above its 200-day average of $71,076. So the account stays in.
Experiments, side by side
Same trades, different rules.
Every page started at $100. The controls copy the council’s trades and change only one setting; the AI pages trade their own signals. So the gap between two lines is that one difference, nothing else.
- The council$39.12
- Control · busy markets only$46.41
- Control · US hours after a big move$108.20
- $108.20
1Control · US hours after a big move
same trades, weekdays 12-18 UTC after a 1%+ move
83% won · 52 trades
- $99.86
2Axiom
trades only when Oracle AI and Wisdom AI agree
57% won · 46 trades
- $95.06
3Noesis
news and principles in one AI judgement
52% won · 155 trades
- $91.18
4Sophos
trades only when three AIs agree
48% won · 31 trades
- $84.04
5Wisdom AI v2
the same, reviewing its own record
48% won · 136 trades
- $80.26
6Oracle AI
follows the AI's reading of the news
47% won · 85 trades
- $70.55
7Noesis v2
the same, reviewing its own record
48% won · 113 trades
- $67.20
8Council + Wisdom v2 + Noesis
council trades neither objects to
45% won · 135 trades
- $62.09
9Council + Wisdom AI
council trades Wisdom AI doesn't object to
48% won · 221 trades
- $55.93
10Wisdom AI
trading principles applied to the market
46% won · 167 trades
- $52.64
11Council + Noesis
council trades Noesis doesn't object to
44% won · 171 trades
- $46.41
12Control · busy markets only
same trades, small target, skips quiet markets
69% won · 232 trades
- $39.12
13The council
the AI council's own trades
49% won · 420 trades
Watch-only strategies
Ideas on trial, in the open.
Each one tests a different way to make money, with fixed rules decided before it started. None is proven. Each shows its live result next to what its backtest promised.
Crash bounce
Buys a coin after a sharp 24-hour drop with a volume spike, sells 4 hours later.
$100.00
Waiting for a sell-off. Closest now: SOL down 2% (fires at down 8.27%)
How it works
The first idea from the AI research engine that passed every check. Since then the engine re-judges live rules against a bar that rises with every idea tried, and this version now falls just short (t 3.4 against 4.2). It keeps trading here because only new data can settle it. $25 per coin, 4-hour hold, fees included.
BTC 2x trend
Holds twice its balance in BTC while BTC is above its 200-day average, cash otherwise.
$99.59
Holding 2x · -1.3% from its peak
How it works
Shows what leverage does, drops included. Backtest: 2024-26 +17.9%/yr against +17.5% for plain holding, worst drop -57%; 2020-23 +53.5%/yr, worst drop -92%.
Selling BTC options
Sells a one-week BTC straddle each Friday, hedged, when options look expensive.
$0.00
Week of 2026-09-25: skipped
How it works
Options usually price bigger moves than happen. On 284 real weeks this made +0.26% a week in 2024-26 (worst week -5.2%), but with t 1.33 it could still be luck. Per 0.1 BTC per leg, Deribit prices, fees included.
Coinbase premium
Follows a sudden US buying or selling spike on Coinbase for 24 hours.
$100.00
Open: short
How it works
Before 2024 this made +50.4 bps a trade (t 3.56); since 2024 +14.9 (t 1.27), same direction but too weak to call proven.
Evolving council
Breeds a new way to weigh the council's 28 agents every day, judged only on the next day.
+1.05 bps
per call · council -9.62 · always buying 11.8
How it works
Promotable only if it beats both the council and simply always buying, over at least 7 unseen days, with a t-score above 2. It never trades.
The AI research engine
Thousands of ideas a day. Almost all of them die.
An AI invents trading rules, a breeder evolves them, and THE CORE’s proof gate judges each on years of data it has never seen, across five worlds: crypto majors, the top 60 coins, every coin ever listed, order-book and news data, and stock and gold contracts.
Ideas judged
90,221
Today
77,343
Live candidates
3
groups of rules, one pattern each
Proof bar (t-score)
5.01
rises with every idea tried
Why ideas died
- Loses money after fees29,481
- Stopped working on new data19,836
- Too rare to judge17,780
- Could be luck among everything tried14,342
- Fails with real-world costs2,787
- No better than doing nothing clever2,124
- One lucky day, not a pattern1,983
The AI council
Can it call direction? A public trial says so far: no.
A council of AI agents debates every candidate trade. On 14 Sep it went on a four-week public trial with the bar fixed in advance: right at least 52% of the time, with a t-score of 2. It now trades its own paper ledger, so its results never touch the main account.
Fair trial · 14 Sep to 12 Oct
49.9%
right on 7,070 calls · t-score 1.99
Needs 52% and t 2. 14 days left. If it fails, the council stays as the live AI demo and stops being expected to predict direction.
Right, day by day
- 51%09-22
- 52%09-23
- 54%09-24
- 48%09-25
- 41%09-26
- 52%09-27
- 51%09-28
The dashed line is 50%, a coin flip. Green days beat it, red days fell short.
The last 10 minutes
Ideas evaluated
569
56.9 a minute
Real AI reasoning
100%
no scripted fallbacks
Approved
48
cleared the 70% bar
Placed
0
after every filter
- Not convincing enough for the council522
- Past results say it loses after fees48
Under the hood
The plumbing, also in public.
Open positions on the main account
- BTC/USD long$168.02 at $84,008−$0.39
Leverage in use
1.68×
Fall to liquidation
59%
how far prices would have to move
Exchange uptime
99.98%
88 ms typical
Liquidations since reset
0
Market data sources
- binance futures102 ms
- binance spot111 ms
- coinbase286 ms
- coingecko269 ms
- frankfurter70 ms
- twelvedatadown
The trading engine reads real prices from the live Binance futures book and books every order on paper, with Binance’s real fees, funding and liquidation rules. The previous version of this page is still at /trading-classic.