Messages lost on producer timeout

Finance reconciliation finds ~0.05% of payment-captured events never reached the analytics warehouse, gaps with NO matching Kafka record at all (not a consumer gap). No consumer lag, no consumer errors. Dashboards: consumers process every record that exists. The producer side shows a small but nonzero record-error-rate and a buffer-available-bytes that periodically dips to ~0 during traffic peaks; the producer was recently tuned for throughput, acks=1, linger.ms=20, max.in.flight.requests.per.connection=5, retries left at a low value, and delivery.timeout.ms fairly short. Brokers had a brief leader election yesterday during the peak window where the gaps cluster. How do you triage and mitigate?