ONE CONSENSUS ACCOUNT
The team’s chosen trades
—Waiting for balances
- Estimated APY
- —
- History elapsed
- —
Uses the combined vote, then applies position sizing and risk rules. This is the number previously called “Treatment net P&L”.
FOLLOW THE DECISIONS. FOLLOW THE MONEY.
One swarm. Separate accounts.
The flies contribute opinions to the consensus. Their account balances are never pooled into it.
Loading current accounts. No balances have been received.
These are different trading experiments
TRADING HISTORY · CALENDAR ELAPSED—Loading recorded start…
CURRENT SESSION—Checking recovery history…
INDIVIDUAL FLY ACCOUNTS
Each fly’s own experiment
—Waiting for balances
- Estimated APY
- —
- History elapsed
- —
Each fly independently tests its own decisions. The amount above adds those separate experiments for reference; it is not the consensus balance.
See each fly’s account ↓ONE CONTROL ACCOUNT
The comparison strategy
—Waiting for balances
- Estimated APY
- —
- History elapsed
- —
Conventional scouts select trades for a separate benchmark. This provides a comparison with the neural consensus.
Loading complete run history before annualizing returns.
How estimated APY and trading duration are calculated
Loading complete run history before annualizing returns.
APY is a compounded annual equivalent of the observed net return, not a forecast or promised yield. Formula: ((current equity ÷ starting equity)365 ÷ elapsed days − 1) × 100. It includes open-position P&L, fees and funding. A short experiment can produce extreme estimates; losses can annualize close to −100%.
We show estimates after 24 hours of recorded calendar history. Duration starts at the original recorded run, includes pauses and downtime, and survives checkpoint recovery. It is not a count of hours actively trading or time since the first fill. All three experiments share this history. Individual flies use their own starting equity; the combined fly estimate uses their total equity and starting capital, not an average of their APYs.
Values use the latest account snapshot and stop advancing when updates stop. Missing or incomplete history leaves APY unavailable. About annualized returns ↗
Each account has its own starting balance. Dollar totals across several accounts are not directly comparable to one account.
WHAT IS SHARED?
Decisions flow across. Money does not.
1
BTC market data
All strategies observe the same market.
2
Fly opinions
Each fly says long, short or hold, while testing its own paper position.
3
Consensus + risk rules
Eligible votes choose a direction. Risk and sizing rules determine whether an order follows.
4
A separate account
Only its own fills, position, fees and funding change the consensus balance.
Waiting for the latest consensus vote.
Hold can keep an existing position open. Muted flies still run their own paper experiments; their votes are excluded from consensus. A current vote is not a fill receipt.
How can every fly lose while consensus wins?
The consensus does not copy every fly’s trade. It can enter at different times, hold for longer and trade less. That creates a different profit history and different fees. A losing private account does not send a bill to the consensus account.
Trade and fee comparisons will appear with the current data.
This separation explains why the numbers can differ. It does not establish that the consensus will keep outperforming.
EVERY ACCOUNT, SIDE BY SIDE
Choose an account to inspect
| Account | Starting equity | Current equity | Net P&L | Return | Est. APY | Fills | Execution |
|---|---|---|---|---|---|---|---|
| Loading accounts… | |||||||
EXPLAIN THIS BALANCE
Swarm consensus
PROPOSED SETUP · EXPLANATION ONLY
Many paper researchers. One real execution account.
Yes: the recommended architecture is to keep every individual fly and the control benchmark on paper, and allow only the consensus strategy to send real BTC orders.
Individual flies
Test ideas, record private P&L and supply votes. Their losses spend no exchange funds.
Consensus + risk limits
Combine eligible opinions into one target position, then check limits before execution.
Hyperliquid BTC account
One strategy owns the exchange position. Its real fills, costs and balances determine actual returns.
What the program already separates
Private fly accounts are explicitly paper-only. The code selects one exchange lane; choosing the consensus leaves the control benchmark on paper too.
Checking the current execution setting…
Reading this page or switching these tabs cannot enable trading or change an account.
What should happen before real funds
- Evaluate the consensus on unseen periods, after realistic costs, against simple baselines.
- Test orders, partial fills, position reconciliation and recovery on Hyperliquid testnet.
- Resolve restart reliability and add exchange-side position protection, with explicit capital and loss limits.
- Use a reviewed, versioned decision policy for an initial small live trial; continue learning in paper and review changes before deployment.
This is a proposed rollout, not a claim that all these protections and promotion controls are implemented. The current system has online readout learning; a paper-to-live model-promotion workflow still needs work.
A positive paper P&L is not enough to declare live readiness. The recent host failure also showed why recovery needs testing. Canceling open orders does not close an already-held BTC position.
Exchange references: Hyperliquid testnet API · Scheduled cancellation cancels open orders.