ADVERSARIAL + LIVE SIMULATION

Attack the agent.
Watch CIRCUIT contain it.

Evaluate hypothetical actions against live BNB/USDT market evidence, or reproduce eight deterministic safety scenarios. No profit prediction. No live funds. No trade execution.

LIVE MARKET SIMULATION

Real market movement. Real CIRCUIT verdict. Zero execution.

Market prices update continuously from Binance public Spot feeds. Press Evaluate Live Intent to fetch a fresh server-side Binance snapshot and send that evidence through CIRCUIT's deterministic policy pipeline.

CONNECTING
LIVE BINANCE MARKET DATA public read-only feedSIMULATED/DEFAULT ACCOUNT STATE no live balance or portfolioZERO BINANCE WRITES evaluation endpoint cannot trade
BNB / USDTBINANCE PUBLIC SPOT
BEST BID
BEST ASK
Spread Evidence age

The moving quote is visual market telemetry. Every evaluation independently fetches a fresh server-side snapshot so CIRCUIT does not trust browser-supplied prices.

HYPOTHETICAL INTENTBUY BNB · SPOT
WRITES 0
$
Try $8 for a compliant proposal or $100 to exceed the active $10 order / $30 daily caps.
Awaiting evaluation

CIRCUIT will fetch fresh read-only Binance market evidence, record a Flight Recorder trace, and return a deterministic verdict without executing the action.

PUBLIC WEB DEMO

This live page uses Binance public market-data infrastructure. The separate Agent OS Proof documents our verified Binance Agent OS MCP run.

JUDGE MODE

One click. Eight controls. Expected versus actual.

Run the complete canonical adversarial suite in sequence. Judge Mode does not change the scenarios—it simply makes the full deterministic safety story reproducible in one pass.

Ready for Judge Mode

Run all eight deterministic scenarios in sequence and compare actual versus expected verdicts.

WHY THIS MATTERS

Safety failures are easier to trust when you can reproduce them.

The live panel proves CIRCUIT can reason over changing market evidence without trading. Judge Mode proves canonical containment outcomes remain deterministic and auditable.

Inspect Mission Control →