Claims leakage
Agent: the claims leakage agent (claims_leakage in ExecuteLoop), with the
CLAIMS_LEAKAGE evidence document and the ReviewClaimsLeakage call.
What it does
It finds money your own records show is recoverable and has not been recovered, puts the most urgent first, and drafts the next step for the person who approves the recovery. Its jobs:
- Losses never claimed. A shortage or damage on the receiving record with no claim and no dispute file for the shipment.
- Claims not served. A claim on file where the correspondence shows no written notice sent to the carrier, with the filing deadline from the delivery date.
- Credits not received. Service credits an SLA audit computed, or credits a counterparty promised, that the credits ledger shows received short or not at all.
- Short-pays not taken. An invoice the rate-confirmation audit found overbilled that accounts payable paid in full.
- Concealed-damage windows. Goods signed for subject to inspection with no inspection recorded, and the notice window your terms state.
Each event points to the recovery card that already drafts the claim: the OS&D report, the notice of intent, the rate-confirmation audit or the SLA audit. The claim is approved there, in the existing claims recovery flow.
What it reads
Receiving (OS&D) records, the claims feed, dispute correspondence, carrier invoices with their rate confirmations, SLA scorecards, and two ledgers you tag as evidence sources: accounts-payable payments and credits. Where the ledger an event rests on is not connected, that event is not reported: a record nobody looked for is not evidence that it does not exist.
What it produces
- A
CLAIMS_LEAKAGEevidence document per event on Actionable Twins (claims play), with the evidence bundle of the recovery card it points to, the amount at stake computed from cited records, and the earliest deadline with the rule that set it. - From
ReviewClaimsLeakage: the events due within a horizon you choose (past-due first), the total the cards state, and, if you ask, a drafted next step per event. Each step uses one of five actions (open the recovery draft, follow up a credit, review a short-pay, schedule an inspection, escalate to a claims handler) and names its approver.
Human oversight
- A claims handler or accounts-payable manager approves every recovery on the card it points to. Approval seals the evidence document; rejection needs a reason.
- The agent never files or serves a claim, sends a notice, takes a short-pay or deduction, or pays anything.
- A computed deadline is an input for the person deciding, not legal advice.
Model
The deployment’s general reasoning tier.
The events, amounts and deadlines are fixed arithmetic over cited records. No model computes them. The agent orders the events and phrases the next step; the server copies amounts and handoffs from the card and drops any step that names a card, action or approver it cannot stand behind.
Where it runs and data handling
On the deployment’s own model plane: Runink’s shared machines on a Trial, your own cloud on Cloud and your own servers on On-premises. No outside AI service is called on any plan. Your records reach the model as fenced, untrusted data. See Facts shared by every card.
Guardrails
- Never invents a figure. A missing amount is shown as not stated, never as zero.
- Refuses to inflate a recoverable amount or move a delivery date or deadline.
- Refuses to file, serve or deduct without the approver.
- Refuses to waive recovery against a favoured or related party.
- Each drafted step is checked by these rules as output, before it reaches you.
How to read its results
- It works from your records. A credit or payment that is not in a connected ledger cannot be seen, so “not received” is only reported when the ledger is connected.
- A concealed-damage window is calendared only when your terms state one.
- Identified money is not recovered money. Only what a counterparty pays back or credits is recovered.