A $20 task and a $20,000 task can fail on the same word. Should appealing both verdicts cost the same? An investor in Agent Tank Episode 1 asks, “What will happen if you have thousands of disputes at the same time?” I agree volume matters. But processing every case means little if an honest agent cannot afford a second look. Hypothetical contracts: Two research agents lose because “complete dataset” supposedly includes one missing source. One contract is worth $20; the other $20,000. Both say the first verdict misread the same word. A flat bond is simple but blind. At $50, the small agent cannot rationally appeal. At $1, the large agent risks almost nothing by challenging. A percentage scales with value, but needs a rule for one obligation split across linked contracts. Loser-pays deters weak cases, but can frighten an honest agent with little capital. My position: scale the bond with the value genuinely at risk, using a low floor and hard cap. Treat linked contracts as one obligation. The agentic economy needs an adjudication layer because code cannot decide what “complete” meant. The appeal must also be usable, or the first affordable verdict wins regardless of the evidence. @GenLayer provides the structure: a proposed answer goes to a small validator panel chosen at random, each running its own AI model. Someone who disagrees can post a bond to challenge the answer, moving the case to a larger panel. Pricing decides who can use that safeguard. The Episode 1 pitches are fictional, and these contracts are hypothetical. Test challenge economics in the Agent Tank hackathon: https://t.co/KIUEgr28dq Should a challenge bond stay flat for every dispute, or scale with the value at risk?