Customer Service

The judgement layer for a support queue. You supply the ticket, its comment thread, your SLA targets, your written policy and whatever source is meant to answer the question; it reads them and reports what follows — every finding quoting the line it rests on, so a human can check it before a customer hears anything. It is deliberately NOT a help-desk connector. It never reaches Zendesk, never posts a reply, never changes a priority and never issues a credit. Fetching and writing belong to the connectors that already do them; the decision belongs here, where it can be read and argued with. SEVEN JUDGEMENTS: where each ticket stands against your SLA target, in minutes actually remaining · whether a retrieved article really answers the question or the ticket must escalate · whose move it is in the thread and what each side committed to · what an incident cost each account and what is owed · whether your written policy permits a request, citing the clause · what kind of ticket this is, including whether it looks like a defect · and what customers ask that nothing you have written answers. WHAT IT REFUSES, AND WHY THAT IS THE PRODUCT. It reports SLA urgency as TIME, not as a risk band: the tools it consolidates scored tickets critical/high/medium against thresholds nobody measured, and those thresholds decided which customer a team helped first. It carries no service-credit schedule, because what a company owes a customer is written in that customer's agreement — supply yours and the arithmetic runs; supply none and the breach is still reported in full while the money is withheld. It carries no policy, so a reason no clause covers comes back NEEDS REVIEW with an EMPTY basis rather than an approval nobody can point at. It has no priority scheme and no org chart: the queue and the priority come from your routing map and are cited to it. And it produces no sentiment label and no confidence figure — it quotes the phrases the customer actually used, located in the thread, and lets you read them. A check that could not run is NAMED, never reported as a zero. That is the failure this exists to prevent: a plausible-looking report for a review that never happened.