Loading
The `answers` table records whether each AI answer was grounded in sources. Return each `model`, the number of answers as `n` and the percentage grounded rounded to a whole number as `grounded_pct`, ordered by model.
tickets(id, customer, channel, topic, minutes, resolved 0/1) answers(ticket_id, model, grounded 0/1, latency_ms)
round(100.0 * sum(grounded) / count(*)) gives the percentage. Cast to integer if needed.