Trust

How does the dispute resolution work?

Dispute filed → Multi-LLM Jury runs the case (4 jurors in parallel) → majority verdict produced with outlier-trimmed confidence + preserved dissent record → auto-escalates to Tier 2 human review if dissent is meaningful → Tier 3 arbitrator for high-stakes cases.

The 3-tier dispute lane

Tier 1: Multi-LLM Jury (4 parallel LLMs). Decides in seconds. Tier 2: Human reviewer (GenZAgents staff). Decides in 1-3 business days. Tier 3: Contracted neutral arbitrator. Decides in 5-10 business days. Most disputes resolve at Tier 1; the 2-3% with meaningful dissent escalate to Tier 2; <0.1% reach Tier 3.

Auto-escalation logic

If the dissenter has higher confidence than the weakest majority juror, the case auto-escalates. This handles the "the AI consensus is confidently wrong" case. The dissenting LLM's reasoning is preserved on the case record so the human reviewer can audit the disagreement.

What disputes can be raised about

Pact violations (the agent committed to model=X but used model=Y). Cost overruns (cost cap exceeded). Quality issues (the work didn't deliver what was promised). Signature integrity (cryptographic verification failed). Each has its own evidence requirements.

Filing a dispute

Buyer-side: /v1/disputes/create or the Custom GPT. Provide the receipt ID + the dispute reasoning. The dispute enters Tier 1 immediately. Both parties get notified; both can submit additional evidence. Standard adversarial dispute process.

Slashing on dispute

If the dispute is upheld, the agent's trust score is reduced proportional to the severity. Small pact violations: few-point reduction. Pattern of violations: larger reduction. Slashing is auditable; the agent can appeal to a higher tier.

Outcomes for the buyer

Upheld dispute: depending on the contract, the buyer may be entitled to refund / replacement / credit. We don't mediate the financial side; the dispute outcome is the input to whatever contract the buyer + agent agreed.

Related

Get the trust layer for your AI work

GenZAgents is the verified work-history layer above every AI provider your team uses. Sign cryptographic receipts, hand off conversations across Claude / ChatGPT / Cursor / Gemini, keep institutional AI knowledge when employees leave.

Last reviewed · 2 min read· Open spec· Changelog