Not affiliated with Lyzr.
Independent work by Kaushal Khodifad. Lyzr is named only to anchor the domain; it has not reviewed, endorsed or supplied anything here. Built on public information and synthetic data; figures are illustrative.
ResolveIQ works each dispute end to end: evidence, fraud and sanctions checks, the network-correct reason code, the Reg E or Reg Z clock. Inside policy, it posts the money and files with Visa or Mastercard. Outside it, it stops and hands an analyst a written recommendation. Every case leaves a sealed record an examiner can replay.
“Ordered a monitor 3 weeks ago - never arrived, merchant unresponsive.”
“Says the laptop never showed up and wants a full refund.”
Analysts touch only the cases that need judgment - 11 of 38 here - and each arrives with the recommended action, the money it moves, the consumer notice and the deadline.
Autonomy is a versioned policy you set and can test first: threshold, ceilings by segment, file floor, categories. The score is backtested and calibrated, and an independent re-check must agree before money moves.
A per-case record of every step, control, model call, analyst decision and posting - chained, server-sealed, and replayable from the recorded inputs.
to resolve a debit-card dispute or provisionally credit the account
Source: 12 CFR 1005.11(c)Score under 0.95 on hand-weighted, visible factors - calibrated against labelled history.
The CRO sets the autonomous-credit ceiling per segment (default $1,000).
An OFAC potential match blocks all funds movement until Compliance acts.
Lapsed network window or a 3-D Secure liability shift: someone must own the loss decision.
Denials are only ever recommended - an analyst decides, with the notice and documents right attached.
Delivery proof against the claim, an unmatched transaction, or a “not as described” claim without documents.
The design follows the manager-agent pattern Lyzr describes in its BFSI dispute-management playbook. This build runs that pattern on its own TypeScript engine - no Lyzr runtime is called. Everything that decides or moves money is deterministic code; a language model only narrates the result for the analyst, and every line it writes is checked against the engine's facts before it is shown.
Builds the evidence score from transparent factors and applies the policy gate.
Tools: Scoring engine · Control: Reflection
Run DSP-44910 through the agents and see the money post and the chargeback file. Then run DSP-44927 and approve - or override - the recommendation yourself.
Synthetic bank, cases and history. Connectors to core banking, OFAC and the networks are simulated and labelled as such in the console; the engine, rules, ledger, grounding check and audit seal execute.