Should I use Jev for SLA deadlines and date checks?
Verdict for the sample idea
Run it on
plain code
The rule can be written down. A function is cheaper and more reliable than any model.
Sample idea
For every open support ticket, decide if we have broken the SLA: compare the time of the customer's last message with the current time, consider business hours and the customer's plan, and flag tickets that are overdue.
Explanation
A deadline check compares two timestamps under a set of rules. That is date arithmetic, and it belongs in plain code. TypeSafe's page on the model's weak points says that Jev reads dates as text and compares them badly. Business hours, time zones, public holidays and daylight saving make the problem harder for a model. For a date library they are routine.
Wrong answers here cost money. An SLA breach that nobody flags becomes a contract penalty. A false alarm at 03:00 wakes a person. You need an exact answer, and you can have one for free.
Jev can still help around the clock, in two places.
The first is the input to the rule. Which SLA applies often depends on a judgment. Is this message a new incident, or a reply on an old one? Is it a severity 1 outage or a question? Does the customer's "thanks, that fixed it" close the ticket and stop the clock? Those are Choice and Noul questions on text. Jev answers them, and code then picks the correct deadline and does the arithmetic.
The second is when a date is hidden in free text. "We need this before the end of next week" contains a deadline. Do not ask Jev for the date. TypeSafe's advice is to extract date parts as choices and to compute in code. In many cases an LLM or a date parser is the better tool for that step.
The common trap is the mixed question. "Is this ticket overdue and urgent?" joins a computation and a judgment in one Noul, so you cannot test either half. Split it. Code answers "overdue", Jev answers "urgent", and code combines the two.
Related tasks
Verdict produced by Jev on 2026-09-19, rubric version 2. Text written by a person. How the tool works: the report.