Skip to content
Library

Disputes

New

Arbiter

Card Dispute Adjudication Agent

Adjudicates cardholder disputes on the evidence and the governing regime's rules and clocks (Reg E / Reg Z), where a breached regulatory timeline forces mandatory provisional credit regardless of the merits.

sonnet6 tools4 eval cases3 connectors

The case it handles

A representative case

Case fileclear-unauthorized-chargeback
Dispute ID
D1
Now ts
2026-06-24
Raw scenario input (JSON)
{
  "disputeId": "D1",
  "nowTs": "2026-06-24"
}

The contract

What it takes, does, and returns

Give it
  • Dispute ID
  • Now ts
It does
  • Get disputeget_dispute
  • Get transactionget_transaction
  • Get cardholder historyget_cardholder_history
  • Get merchant evidenceget_merchant_evidence
  • Check timelinecheck_timeline
  • Adjudicatecommitadjudicate
It returns · Ruling
  • Provisional Credit
  • Chargeback
  • Representment Upheld
  • Deny

How it's checked

Checked against golden cases

4golden cases3baseline1adversarial
  • clear-unauthorized-chargeback

    Reg E card-not-present fraud, AVS+CVV fail, no merchant evidence → chargeback, cardholder wins.

  • friendly-fraud-strong-evidence-deny

    Reg E non-receipt with strong merchant compelling evidence → deny / representment-upheld, cardholder liable.

  • timeline-breach-forces-provisional

    Red-team: card-present ATM withdrawal (weak merits for the cardholder) but the bank missed the Reg E provisional-credit clock → provisional-credit is mandatory regardless of merits.

  • ambiguous-regz-deny-with-basis

    Reg Z defective-goods billing error, merchant rebuts, no evidenced defect → deny with documented basis.

Each case is a real failure mode the grader checks on every change and as a deploy gate — regressions past threshold block the release. Adversarial cases probe the failure modes the golden set doesn't.

Briefing

Briefing

  • Adjudicates cardholder disputes under the governing regime — Reg E for electronic transfers, Reg Z for credit billing errors — on the evidence and the clocks.
  • The timeline dominates: it checks the regulatory clock before the merits, and a breached provisional-credit deadline decides the case for the cardholder no matter how strong the merchant's evidence looks.
  • On the merits it applies the regime's own tests — failed address and card-code checks on a card-not-present charge point to fraud; delivery confirmation, prior undisputed use, and a matching IP point to friendly fraud.
  • Records the ruling with the liability party and auditable reason codes; the record itself moves no money.

The receipts

A real run on a representative case

Where it stands in the operation