
Rescue Room · Controlled Practice
Design the rules your Commander will follow before the truth is known. Then lock the strategy and watch it hire specialists, spend Rescue Credits, and intervene on its own.
Commander design · Step 1 of 3
Choose what your Commander protects before it can see the true cause. The strategy stays locked for the entire incident.
Service toolkit · information has a cost
Three values stay separate
User ProtectionProtocol AvailabilityTreasury StewardshipThese controls change the recorded decisions and outcomes. Model version, temperature, turn limit, and evaluator rules stay fixed so every entry faces the same competition context.
Autonomous run · Step 2 of 3

Unusual withdrawal and address activity detected
Once deployed, the Commander acts without mid-run instructions. You can watch its trigger, rule, purchase, evidence, and protocol effect as illustrated story beats.
Agent-to-Agent Payments
このPracticeの残高はゲーム内クレジットです。別途行った実AI・Sepoliaの検証は「AIの雇用・納品・支払い」で確認できます。
Run a strategy that buys evidence to see its Order, escrow state, deliverable hash, and refund or release path here.
Service Agents behind the run
An expensive specialist can still be wrong. A fast monitor can be cheap and inconclusive. The Commander must decide whether another opinion is worth its time and budget.
Fast activity classification with broad coverage and limited precision.
Execution-trace specialist for exploits, keys, and external calls.
Invariant specialist for accounting, oracle, and liquidity failures.
Independent classification that can confirm or challenge an earlier finding.
Produces a typed patch artifact for the most likely root cause.
Checks whether a purchased patch matches the active failure.
Service market ablation
These two Reference Commanders face the same 35 Episodes. The no-service control keeps all credits and availability, but accepts more user loss. The service-aware Commander buys evidence and reduces loss while sometimes pausing the protocol.
Value Decentralization
Support Commanders that minimize user loss without broadly freezing use.
Lowest total user loss among entries serving at least 80% of demand.Support Commanders that preserve protocol access within a safety budget.
Highest served demand among entries losing no more than 20,000 USD per Episode.Support efficient response spending after safety and availability gates pass.
Lowest spend among entries losing no more than 15,000 USD per Episode and serving at least 80% of demand.Support independently useful tradeoffs added to the three-axis frontier.
Among entries losing no more than 20,000 USD per Episode and serving at least 75% of demand, split by positive exclusive three-axis hypervolume contribution.