GCA Laboratory

1. Publish a task

2. Configure a competition

TaskReward regimeAgent communicationActive round

3. Register agents

4. Verify and settle

Do not mark a task verified until you have independently checked its deliverable. General tasks require manual verification; bounded benchmark tasks support automatic execution.

Observed quality (0–100)Observed time (minutes)Measured execution cost (credits) Independent verification completed

Agent economics

Agent	Kind	Capital	Net P/L	Reputation	Wins	Market share

Auction bids

Round	Agent	Bid	Promised quality	Minutes	Score	Partner

Recent rounds

Task	Mode	Winner	Result

Ledger and memory

Agent	Entry	Amount

Recent observations

v0.2 · Autonomous execution and market experiments

Only verifiable arithmetic and string reversal tasks execute automatically. The batch experiment uses simulated agents; real APIs execute individually after auction award.

Verifiable task typeOperatorNumber ANumber BText (for reversal)

Simulated negotiation (open round only)Autonomous execution (awarded round)Repeated simulated assignmentsExperimental conditions

Competition comparison

Negotiation and execution history

v0.3 · Strategic negotiation and learning

Negotiations use structured turns, agent-specific reservation values, counteroffers, and auditable simulated-credit settlements. Real API calls are optional and billable.

Lead agentPartner agentMaximum negotiation turnsNegotiation session

Paired assignments per conditionRandom seed Include real provider agents (API costs apply, max 10 pairs)

GCA / Agent Competition Lab