Consensus account P&L
—
Separate account ↗ — fills
OBSERVE · COMPARE · IMPROVE
Research overview.
Understand what the swarm is doing—and what the evidence supports.
ⓘ
Connecting to your research desk. No market data has been received yet.
—
Control net P&L
—
Scout ensemble — fills
Consensus minus control ↗
—
Observed difference No significance claim
Consensus drawdown ↘
—
From recorded peak — exposure
Performance comparison
Equity after execution costs and funding
Treatment
Control
USDC · recorded cycles
⌁
Your experiment starts here
The chart will appear after two recorded trading cycles.
Experiment awaiting observationsLearning & evidence ↓
Swarm explorer
Open full explorer ↗Visualization only. Performance and risk remain in the dashboard above.
INDIVIDUALS BEFORE CONSENSUS
Meet the agents
Inspect each private book, decision and learning signal.
| Agent | Decision | Private net P&L | Reputation | Individual credit | Group credit | Voice | Details |
|---|---|---|---|---|---|---|---|
| Waiting for agent observations. | |||||||
KNOW WHAT IS BEING LEARNED
Learning & evidence
A learning signal is a hypothesis. Evaluation is how you test it.
Reward the outcome
Credit follows the policy that held the position, including fees, funding and forced exits.
- Reward basis
- Net equity change
- Observed reward intervals
- —
Keep research separate
Neural replay uses an independent research brain. Search uses earlier completed candles, with the final 20% reserved.
- Training generation
- —
- Automatic neural promotion
- Off
Earn the next step
Compare the complete swarm, frozen learning and simple baselines across reserved windows, with reward ablations and doubled fees.
- Current display
- Forward observation
- Validated trading edge
- Not established
Run recording
Completed cycles, fills and rewards stay together in a durable run history.
- Run ID
- —
- Last checkpoint
- —
- Committed events
- —
Public input archive: idle.
Feed monitoring starts with this process.
Historical training worker: idle.
Download full event history ↓Simulation recovery requires a compatible checkpoint and a fresh feed.
How to interpret these results
Simulation uses aggressive fills with spread, slippage and taker fees. It does not model queue position, market impact or liquidation. Exchange fee totals are estimates until actual fee ingestion is added. Reputation and nectar are learning signals, not confidence probabilities. Use the repository’s full-swarm evaluator for recorded hourly comparisons, input archives and conditional uncertainty estimates. Promotion stays disabled. Repeated tuning against the same test period invalidates its independence. The synthetic mushroom-body motif and optional full connectome are distinct backends.
FROM DECISION TO RECEIPT
Execution log
| Time | Agent / lane | Side | Quantity | Price | Source | Reason |
|---|---|---|---|---|---|---|
| No fills recorded. Orders appear only after execution. | ||||||