Combining with SQL

jev() is a boolean. Put cheap, indexed predicates first so fewer rows reach the model.

sql

SELECT * FROM tickets
WHERE status = 'open'
  AND jev(tickets, 'the customer is angry');

It joins like any other function. The row you pass still needs a real table or view type so the read-ahead can batch it; see How it works.

sql

SELECT c.name, t.subject
FROM tickets t
JOIN customers c ON c.id = t.customer_id
WHERE jev(t, 'this is a billing dispute');

Drop columns the model should not see. A view is easier to reuse than a nested SELECT:

sql

CREATE VIEW ticket_for_jev AS
SELECT id, subject, body FROM tickets;

SELECT * FROM ticket_for_jev v
WHERE jev(v, 'the customer threatens legal action');

GROUP BY works on jev_choice() (or anything else you select):

sql

SELECT jev_choice(tickets, 'which team should handle this?',
                  ARRAY['billing', 'technical', 'security', 'sales']) AS team,
       count(*)
FROM tickets
GROUP BY 1;

Subqueries and CTEs that expose an anonymous record cannot be read ahead and cost one request per row. Prefer a base table or a view. For wide scans, Large tables.